1 Introduction
The intensification of global climate change and the worsening energy crisis have fueled the urgent need for urban energy conservation and the development of ecological green cities. According to the International Energy Agency (IEA), building operations accounted for 36% of global energy consumption in 2018, with space heating alone responsible for approximately 34% of that consumption (
IEA, 2018). In China, urban residential districts contribute 36%–40% of total urban building energy use, especially amid rapid urbanization and the proliferation of highrise residential buildings (
Efficiency, 2020;
Gou et al., 2018). Although buildings in northern cold and severe-cold regions comprise only half of China’s urban building stock, their heating energy use accounts for 40% of national building operational energy consumption (
Deng et al., 2021). This disproportionate energy burden highlights the critical need to improve residential energy efficiency in cold regions through morphological and early-design interventions.
In response to these challenges, a growing body of research has explored factors influencing residential energy consumption from multiple dimensions. Most studies focus on the thermal performance analysis of individual buildings. For example,
Wang et al. (2021) provided valuable energy-saving solutions by comparing the energy consumption, economic performance, and lifecycle performance of enhanced insulation materials and building-integrated photovoltaic systems. The impact of occupant behavior on building energy consumption has also received widespread attention. In one study,
Yan et al. (2023a) systematically reviewed and analyzed how occupant behavior impacted building energy consumption and thermal comfort by exploring various modeling methods and strategy solutions. The results showed that thermal comfort (i.e., environmental conditions) triggers occupants’ behavior toward corresponding building systems and results in energy-load changes, thereby impacting building energy consumption. Regarding the efficiency of indoor equipment and systems,
Yelisetti et al. (2022) explored the role of home energy management systems (HEMS) in energy conservation, finding that monitoring energy consumption and scheduling device operation can significantly reduce energy costs.
Li et al. (2021) compared the microclimate conditions of existing residential districts with those under centralized heating systems. In the study, the researchers examined how urban morphology parameters such as coverage ratio, average building height, and floor area ratio impacted heating energy efficiency in residential districts. The findings revealed that, compared to low-density building areas, high-density building areas have a more significant impact on outdoor air temperature at night through centralized heating systems. Collectively, these studies provide valuable insights into factors affecting building energy consumption, but residential energy consumption is not only related to individual buildings but also influenced by the overall spatial morphology of the residential district. The overall research in this area is mostly qualitative and fails to provide a comprehensive understanding of practical applications.
With the growing demand for building design and energy conservation, simulation-based parametric optimization methods are gaining increasing attention to improve overall building performance. For example,
Baghoolizadeh et al. (2022) used photovoltaic shading devices for multi-objective optimization (MOO) to reduce annual electricity consumption and increase electricity production. The study simulated a three-story building in Tehran using EnergyPlus software, demonstrating significant energy-saving effects under idealized conditions. Similarly,
Araújo et al. (2023) introduced an efficient MOO method using surrogate models to minimize the energy use and construction costs of residential building clusters. By employing machine learning and optimization techniques, the researchers created a surrogate model and conducted simulations through virtual models. The results showed significant improvements in optimization results.
Wu et al. (2024) combined the BO-XG Boost and NSGA-II (non-dominated sorting genetic algorithm II) frameworks to optimize the energy consumption, thermal comfort, and lighting performance of residential buildings. Using the Grasshopper platform for simulations and Latin hypercube sampling (LHS) for dataset acquisition, the method employed in the study was shown to significantly reduce energy consumption and enhance residential performance. Meanwhile,
Xue et al. (2021) investigated how to minimize the lifecycle costs and CO
2 emissions of passive residential buildings under severe cold-climate conditions, using variables such as building materials, window-to-wall ratio, and building orientation. Despite showing the overall theoretical potential of MOO, these studies encounter numerous challenges in practical application. Current research is mostly based on idealized models and therefore lacks the comprehensive consideration of market demand, building codes, and actual design processes. Additionally, previous studies have not systematically considered design variable settings from the early stages of planning, such as plot ratio, residential layout types, and the proportion of highrise buildings. The studies have also paid limited attention to the spatial morphology of residential districts, which is essential for optimizing overall energy efficiency and performance.
In China, a key step in the early stages of residential project planning includes determining the floor area ratio (FAR). The FAR, which represents the relationship between the total building area and the land area, directly impacts the economic viability and feasibility of the project. To ensure project profitability, developers usually maximize the FAR within the allowable range. This process requires not only considering the relevant policies and regulations, but also ensuring a balance between building density and environmental quality. Therefore, in practical residential project planning, setting a reasonable FAR and optimizing building morphology are important strategies for achieving energy savings and improving living quality. By optimizing the overall spatial morphology of residential districts, energy consumption can be significantly reduced and resident living quality can be improved, promoting urban sustainability. This study aims to fill this gap in the research related to energy efficiency in cold-climate regions and bridge the divide between theory and practice. By integrating actual project surveys, market demands, and building codes, we explore optimization methods for residential district spatial morphology based on common FAR and land sizes in cold-climate regions, using Qingdao as a representative case. Our systematic analysis of residential district spatial characteristics comprehensively examines their impact on building performance. This study provides new perspectives and strategies for improving energy efficiency, enhancing indoor comfort, and optimizing daylight utilization in residential buildings within cold-climate regions.
Ultimately, this study aims to address how to design efficient and practical residential district planning schemes while maximizing the FAR and how to optimize the spatial morphology of residential districts to enhance energy efficiency and living comfort. Unlike traditional idealized models, this study is grounded in the early stages of project planning and proposes several typical residential district planning models. We administered surveys on the current situation and selected common FAR and land sizes to conduct detailed optimization research on these models while ensuring maximum FAR, aiming to provide planning strategies better aligned with practical application needs. To that end, this paper proposes a method based on the MOO tools Grasshopper and Wallacei to systematically analyze how residential district spatial characteristics in cold-climate cities comprehensively impact energy consumption, thermal comfort, and daylighting performance. We established six models with different building forms based on surveys and market analysis of typical residential districts in Qingdao, which serves as a representative city for cold-climate regions. Using variables such as building orientation, floor height, window-to-wall ratio, and building form type, we conducted MOO on energy consumption, natural lighting, and thermal comfort, and based on the results, proposed energy-saving residential district planning strategies are suitable for cold-climate regions.
The contribution of this study lies in the development of a validated, transferable MOO framework that integrates parametric modeling, large-scale simulation, and statistical analysis within an early-stage planning context. The six residential models were derived from 660 real-world residential districts in Qingdao and strictly follow local planning codes, ensuring strong engineering relevance. In contrast to prior MOO studies―which often focus on individual buildings, assume simplified input conditions, or neglect practical design variables―this study incorporates realistic site constraints and urban morphological settings. Specifically, it addresses critical research gaps by (1) introducing plan layout and height distribution as early-phase design variables, (2) combining Pareto-based optimization with non-parametric statistical verification to enhance robustness, and (3) operating at the residential district scale to bridge the disconnect between academic methods and real-world urban planning. These contributions provide both theoretical advances and practical strategies for energy-efficient design in cold-climate regions.
2 Literature review
2.1 Energy consumption in cold-climate regions
According to the climate zoning in China, cold-climate regions include Tianjin; Shandong; Ningxia; most parts of Beijing; Hebei; Shanxi; Shaanxi; southern Liaoning; central and eastern Gansu; and parts of northern Henan, Anhui, and Jiangsu. The main climatic characteristics include average January temperatures ranging from –10 ºC to 0 ºC and average July temperatures between 18 ºC and 28 ºC. In these regions, the winters have short daylight hours and low sun angles, and the summers are relatively hot and humid with concentrated rainfall. The large annual temperature variation also poses significant challenges for building energy conservation (
Huang et al., 2020). For winter heating and summer cooling, energy consumption is substantial in cold-climate regions, indicating significant energy-saving potential. Additionally, cold-climate regions are major population centers in China, so energy conservation in residential areas is critical.
2.2 Building morphology and energy consumption
The geometric shape and density indicators of buildings significantly affect energy consumption. Common building morphology indicators include building density (BD), floor area ratio (FAR), building coverage ratio (BCR), sky view factor (SVF), wall surface area (WSA), floor area (BF), open space ratio (OSR), shape coefficient (SC), and window-to-wall ratio (WWR). Using Jianhu, China, as an example,
Liu et al. (2023b) extracted building types from the actual urban context to create blocks and explore the relationship between building morphology and energy consumption. The researchers found that the building density, open space ratio, shape coefficient, and average perimeter ratio were correlated with the total energy use intensity of the blocks.
Leng et al. (2020) studied 73 office buildings in Harbin, a city in China’s severe cold-climate region, to evaluate how seven urban morphology factors impact heating energy consumption. The results showed that greater building site cover, floor area ratio, building height, road height-width ratio, total wall surface area, and lower green space ratio help reduce heating energy consumption. Among these metrics, FAR is the most critical factor, with each unit increase saving up to 10.820 kW h/m
2/year of heating energy.
Other studies have examined how different aspects of urban morphology influence building energy consumption and efficiency.
Mangan et al. (2021) explored how urban morphology impacts building energy consumption and cost-efficiency. By analyzing different urban configurations, the study found that the building-height-to-street-width (H/W) ratio plays a significant role in energy consumption and economic performance, while the impact of building orientation is minor.
Taleghani et al. (2013) evaluated layouts of point, slab, and courtyard buildings in the temperate climate city of Rotterdam, finding that courtyard buildings have the lowest thermal energy demand in summer and the longest thermal comfort periods. In another study,
Liu et al. (2020) analyzed how urban densification impacts building energy consumption. The study showed that, in China’s hot summer and warm winter regions, dense building layouts reduce cooling loads by decreasing the absorption of solar and longwave radiation.
Through dynamic energy consumption simulations and global sensitivity analysis, scholars have also discovered that building layouts considerably affect cooling, heating, and total energy consumption (
Martins et al., 2016;
Vartholomaios, 2017). The energy consumption of residential areas is not only related to individual buildings but is also influenced by the morphological characteristics of the group (
Quan et al., 2016). Specifically, the mutual shading between buildings and microclimate effects make the energy consumption different due to the addition of individual building energy consumption (
Pisello et al., 2012). Numerous studies have shown that optimizing group morphology design can effectively reduce overall building energy consumption (
Salat, 2010;
Wong et al., 2011).
Overall, the analysis of existing studies revealed that, while the optimization of individual buildings plays an important role in building energy consumption studies, research from the perspective of group morphology is gaining increasing attention. Through the optimization of the spatial morphology of building clusters, energy consumption can be more effectively reduced, improving energy efficiency and providing important references for sustainable development.
2.3 Application of multi-objective optimization in residential district planning
To achieve optimal building performance and meet residents’ needs, designers must make trade-off decisions among multiple objectives. MOO is widely adopted to simultaneously enhance multiple performance objectives in design. As a global search technique, genetic algorithms (GAs) can find multiple Pareto-optimal solutions and are an efficient tool for solving MOO problems (
Diakaki et al., 2008). The objectives of MOO typically include improving indoor thermal comfort, reducing building energy demand, and minimizing environmental impact.
Optimization involves selecting the best elements from available alternatives to find the minimum or maximum value of the objective function, considering multiple constraints. Typically, the objective function is calculated using scientific simulation tools (
Machairas et al., 2014). Simulation-based research has advantages such as low cost, broad applicability, and high reliability compared to empirical research (
Lee and Jeong, 2018;
Oh et al., 2021). In addition, Ladybug tools are gaining increasing attention in energy assessments due to the ability to combine the profits of parametric tools in building and urban geometric modeling (
Natanian and Auer, 2020).
In the field of architecture science, however, the most common MOO method is the Pareto dominance approach, which can optimize all objectives simultaneously and provide a set of non-dominated optimal solutions for decision-makers to choose the best plan (
Asadi et al., 2014;
Gossard et al., 2013). The NSGA-II algorithm developed by
Deb (2011) is recognized as an effective algorithm for MOO and has been widely applied in the field of building performance optimization.
In Chinese residential buildings, the demand for special morphology design is also increasing, requiring the design of high-performance buildings that meet multiple sustainability standards. However, optimizing highrise residential buildings often involves conflicting objectives, such as thermal comfort, lighting, and energy demands for cooling and heating (
Chen and Yang, 2017;
Wang et al., 2021b). MOO methods can effectively address these design problems. In applications,
Toutou et al. (2018) employed the Octopus plugin, a genetic algorithm optimization tool based on Grasshopper, to optimize parameters such as WWR, building materials, and shading for residential buildings to achieve optimal lighting and energy performance.
Fialho et al. (2011) focused on optimizing building materials and orientation to reduce building costs and energy use.
Gou et al. (2018), utilizing the NSGA-II algorithm, improved annual indoor thermal comfort and reduced annual building energy demand, thereby finding optimal solutions for new residential buildings in Shanghai.
Collectively, these studies and application examples show that the wide application of MOO methods in residential district planning not only improves building performance but also provides scientific decision support for designers, promoting the sustainable development of energy-saving cities.
2.4 Trends in residential district development
Since 2000, urban residential construction in China has entered a period of rapid development, with the emergence of midrise, mixed highrise, and highrise residential districts. The FAR has also gradually increased, making highrise residential buildings the dominant form, particularly in major cities such as Beijing, Shanghai, and Guangzhou. The proliferation of highrise residential buildings is both an efficient use of urban land resources and a necessary means to meet the growing urban population demand. However, high-density energy consumption and limited space for integrating renewable energy pose significant challenges to low-carbon transitions that often result in high cooling and heating demands. Therefore, optimizing design to reduce energy consumption has become a key research focus (
Wang et al., 2020).
To address the challenges of energy consumption and carbon emissions in highrise residential buildings,
Pan and Ning (2014) applied the socio-technical systems (STS) theory to offer new perspectives and solutions. The study indicated that achieving zero-carbon highrise buildings requires technological innovation and changes in social behavior and culture. Strategies such as reducing demand, improving energy efficiency, integrating renewable energy, and implementing carbon offsets can lower energy consumption and emissions.
The design and optimization of highrise residential buildings involve multiple aspects, with increasing attention to the quality of the internal environment for aspects such as lighting, visual comfort, thermal comfort, and energy consumption. For example, studies have shown that appropriate lighting design can significantly improve residents’ psychological and physiological states (
Wang et al., 2023).
Elgheriani and Cody (2019) highlighted how the design parameters of highrise residential buildings often lack consideration of the surrounding environment, leading to many environmental problems during construction and operation. Assessing the impact of design parameters on energy and thermal performance can therefore improve the overall performance of buildings.
Future research should systematically compare the impact of different geometric factors (e.g., building orientation and WWR) on the thermal performance and energy consumption of highrise residential buildings in different climate regions. With the continuous advancement of urbanization, highrise residential buildings in China will continue to develop, so finding ways to achieve high-quality living environments in limited urban spaces is an important research direction.
2.5 Summary
Existing research has made significant progress in improving building performance and optimizing energy use, but limitations remain, particularly the reliance on idealized building-scale models, limited consideration of early-stage morphological decisions, and insufficient alignment with actual regulatory and market contexts in cold-climate regions. Accordingly, this study has the following main objectives:
(a) propose a MOO workflow for optimizing high-performance residential morphology grounded in urban planning practice;
(b) use this workflow to identify and analyze high-energy-performance residential forms based on real-world configurations;
(c) quantitatively assess the impact of spatial morphology variables on energy use, daylighting, and thermal comfort using large-scale simulation data;
(d) derive evidence-based strategies for cold-climate residential planning that align with policy constraints and development norms.
Unlike most MOO frameworks that focus on standalone buildings or idealized prototypes, this study synthesizes six residential forms from 660 actual districts in Qingdao (Section 3.1.2), ensuring full compliance with national codes (e.g., GB 50096-2011). It introduces two underexplored morphological variables at the district scale―plan layout and vertical height mix―and evaluates them using a dual-layer framework that combines Pareto optimization and non-parametric statistical testing. These innovations offer enhanced realism, greater methodological rigor, and direct relevance to early-stage design decisions in cold-climate urban contexts.
This work thus contributes both methodologically and practically: it bridges the gap between simulation-based optimization and on-the-ground planning constraints, and offers new insights into the morphology–performance relationship that can guide architects, planners, and policymakers toward more sustainable urban development.
3 Method
This study presents a MOO experiment with a residential district design. Through the design, we evaluated how spatial morphology impacts building energy consumption, natural lighting, and indoor thermal comfort and, based on the results, proposed optimization strategies for cold-climate regions. The research process involved three steps: (a) investigating the spatial morphology characteristics of existing residential districts and constructing parametric baseline models; (b) defining optimization variables and objectives, and establishing a MOO problem; and (c) conducting simulation optimization, screening, and comparing optimization results to derive the optimal Pareto solutions. The overall workflow is shown in Fig. 1.
The study is based on the Grasshopper platform in Rhino, using Ladybug and Honeybee tools for performance simulation, the Wallacei plugin for MOO, and SPSS software for statistical analysis. These tools and methods have been widely adopted in previous peer-reviewed studies on building energy performance and optimization (
Han et al., 2018;
Molake et al., 2023;
Sadeghipour Roudsari and Pak, 2013;
Wang et al., 2021b;
Wu et al., 2024), ensuring methodological reliability and reproducibility. Similar to GA-based daylight-energy optimizations for office spaces in Tehran (
Hakimazari et al., 2024), our NSGA-II workflow prioritizes early-stage variables (orientation, WWR) but extends applicability to cold-climate residential districts by incorporating vertical distribution impacts.
3.1 Current situation survey of residential districts in the study area
This section introduces the characteristics of the study area―including the geography, climate, and residential morphology―and determines the features and scale of the simulated residential districts based on the collected information.
3.1.1 Geography and climate
This study uses cold-climate regions as the primary research subject, with Qingdao as a representative city for case analysis. In China’s building climate classification, Qingdao is in Climate Zone A for cold-climate regions. The winter season lasts from November to March, with the coldest month being January due to its average temperature of –1.5 ºC and the hottest month being August due to its average temperature of 25.1 ºC. Requiring both winter heating and summer cooling, Qingdao represents an ideal city for studying urban residential energy consumption in cold-climate regions. The study area is densely built and has diverse building types, resulting in concentrated urban economic activities and resident life.
3.1.2 Current situation analysis of residential districts
Data on 716 residential districts built since 2010―including construction time, land area, building area, FAR, layout type, and greening rate―were obtained from the
Fangtianxia (2025) database. After the exclusion of samples with missing data, 660 complete samples were obtained. These statistics served as the basis for creating the foundational models.
The statistical data (Fig. 2(a)) show that most samples are below 30,000 m2, with the most common size being 20,001 to 30,000 m2 and accounting for 24.24% of the samples. While 72 samples have a land area of less than 10,000 m2, only seven samples are larger than 90,000 m2. The FAR (Fig. 2(b)) shows that the 1.5 to 2.0 range has the most samples (155). Samples with a FAR above 6.0 and below 1.0 are rare. The spatial layout types (Fig. 2(c)) are mainly categorized into Slab Layout, Point Layout, and Point-Slab Hybrid Layout; Slab Layout was the most common (529), followed by Point Layout (58) and Point-Slab Hybrid Layout (73). Considering the prevalence and potential energy impact of these types in cold-climate regions, this study included these three layout types in the foundational models to analyze the energy consumption impacts of the layouts and provide more options for subsequent optimization in similar regions.
3.2 Model configuration
Based on the survey results, representative site conditions were determined. Six basic models were created on a rectangular site of 24,000 m2 (160 × 150) with a FAR of 1.8, building density not exceeding 20%, and a green space ratio not less than 35%. Five preliminary design variable indicators were defined, and three optimization goals were established as the foundation of the MOO study.
3.2.1 Establishment of typical models
Through extensive surveys of existing building projects in Qingdao, representative samples were selected after an analysis of project scale, building quantity, and morphological diversity to determine the basic characteristics of the models. A site area of 24,000 m
2 with a FAR of 1.8 was chosen, reflecting the common range in the Qingdao residential market. Typological methods (
Kamal et al., 2021;
Li et al., 2021;
Natanian and Wortmann, 2021;
Ratti et al., 2003) widely used to explore forms of residential districts in China were then applied (
Zhang and Gao, 2021;
Zhang et al., 2022). Based on previous research and classifications, the primary classification standards were summarized from two dimensions:
plan layout and
vertical distribution.
3.2.1.1 Plan layout
a) Point Layout: Buildings have compact footprints with a length-to-width ratio ≤2:1 (e.g., 24 m × 12 m). Independent units were scattered to optimize ventilation and natural lighting.
b) Slab Layout: Buildings have elongated footprints with a length-to-width ratio >2:1 (e.g., 52 m × 11.5 m), arranged in strips to improve land-use efficiency and views.
c) Point-Slab Hybrid Layout: Combines features of both layouts, balancing spatial efficiency and environmental quality. It typically involves integrating towers and slabs on the same site.
These distinctions are illustrated in Fig. 3 to clarify geometric differences.
3.2.1.2 Vertical distribution
Given the impact of highrise buildings on residential district development, the classification of vertical distribution focused on the proportion of highrise buildings. Prior studies have indicated notable differences in energy consumption between highrise and midrise buildings, necessitating varied proportions of highrise buildings in district planning. Based on the prior studies, the following classifications were established for this study.
a) Highrise Only: Buildings have heights exceeding 27 m and up to 100 m. Suitable for dense urban areas, effectively utilizing space.
b) Mixed Height (highrise: midrise = 1:1): A combination of high-rise buildings (above 27 m and up to 100 m) and mid-rise buildings (above 10 m and up to 27 m), to enrich the visual hierarchy and residential environment.
Combining these classification standards in pairs through a matrix identified six different residential forms: Point-Highrise, Point-Mixed Height, Slab-Highrise, Slab-Mixed Height, Point-Slab Hybrid-Highrise, and Point-Slab Hybrid-Mixed Height. In early project planning, each form is generally considered as a potential solution by designers and developers. The six classifications for this study reflected the current state of the residential market in Qingdao. These six model combinations were not hypothetical constructs but were derived from statistical analyses of 660 real-world residential districts in Qingdao (see Section 3.1.2), ensuring their representativeness. The spatial dimensions, FAR values, and layout proportions were selected based on aggregated data distributions. All geometric parameters―such as building footprints, site coverage, spacing, and height limits―strictly follow local planning regulations and the Design Code for Residential Buildings (GB 50096-2011) (
MOHURD of China, 2011). This ensures the practical validity and engineering relevance of the proposed form typologies. By simulating the energy consumption and living performance of the different combinations, the impact of design variables and building forms on the energy efficiency, daylighting, and thermal comfort of residential districts could be more accurately assessed.
The simulation site was a rectangle of 160 m × 150 m, adhering to local building codes (
Bureau, 2018): buildings were set back 5 m from the red line, building height was limited to 80 m, spacing between highrise buildings was no less than 13 m, spacing between highrise and midrise buildings was no less than 9 m, and spacing between mid-rise buildings was no less than 6 m. To meet daylighting regulations, the south-facing room windowsill height (0.9 m) in each building must have at least 2 h of sunlight on the winter solstice. For energy consumption and daylighting simulations, the geometry of individual buildings was simplified to rectangular prisms. In this study, all residential buildings were modeled as simplified rectangular prisms. This approach reflects actual early-stage planning practices in China, where schematic designs are typically developed based on rectangular massing to ensure efficient land use and FAR compliance. Detailed articulations such as balconies, setbacks, or façade modulation are introduced in later stages of architectural refinement. Moreover, similar geometric simplifications have been adopted in prior urban-scale optimization studies (e.g.,
Feng et al., 2024;
Liu et al., 2023b), where simplified blocks were used to evaluate energy consumption and solar potential across multiple residential clusters. This abstraction ensures consistency with practical design workflows while maintaining computational efficiency across large-scale parametric simulations. The standard footprints for point residential buildings were 24 m × 12 m and 25 m × 12.5 m (width × depth), with two sizes available. For slab residential buildings, the standard footprint was 52 m × 11.5 m (width × depth). Based on these parameters, six typical residential models were constructed. The models are shown in Table 1, which presents the specific model characteristics.
3.2.2 Definition of variables
In the early stages of architectural design, exploring energy-efficient designs is a crucial step to ensure optimal building performance (
Gou et al., 2018). The spatial morphology of residential district planning directly influences its final performance. However, traditionally, these plans are often based on builders’ experience and personal preferences. Adopting scientific spatial morphology and layout is therefore critical to optimizing building performance. Based on the six typical foundational models defined earlier, this study introduced five key variables: orientation of building, WWR (three variables: north, south, and east and west), and height of bulge. These variables are generally easily adjustable in the initial design stages, providing developers and designers with extensive flexibility and choice. Other potentially influential factors, such as external shading, façade materials, or neighborhood density, were intentionally fixed. These elements are typically addressed during later design development stages, and allowing them to vary would obscure the independent impact of spatial morphology on building performance. Floor area ratio (FAR), site coverage, and envelope parameters were therefore constrained according to local design standards to maintain experimental control. Table 2 presents the specific settings of these variables, providing detailed parameter bases for this study.
Orientation of building (OB): Building orientation helps control and utilize natural light. Appropriate orientation can maximize daylight use and reduce seasonal heat overload.
Window-to-wall ratio (WWR): The WWR determines building daylighting and energy consumption. Larger windows can provide ample natural light but may increase energy consumption.
Height of bulge (HB): Height of bulge affects indoor natural light distribution and ventilation efficiency. Greater floor height helps improve air circulation and light penetration but may lead to increased energy consumption.
To ensure the practicality and robustness of optimization, the step lengths of key variables were carefully determined based on regulatory standards, literature precedents, and physical sensitivity. Specifically, the 0.05 step size for WWR corresponds to China’s modular facade guidelines (GB/T 50002-2013) (
MOHURD of China, 2013b), which adopt 300 mm increments, approximately 5% of a standard 6-m facade width. The 0.1 m step for HB aligns with the recommended vertical module (100 mm) in Technical Specification for Concrete Modular Buildings (
MOHURD of China, 2023) when floor height is below 3.6 m. The 5º step for orientation follows settings used in prior studies on daylight and comfort optimization.
3.2.3 Performance evaluation indicators for residential districts
This study aimed to comprehensively evaluate and optimize residential district design to achieve the optimal balance of energy efficiency, comfort, and daylighting performance. Accordingly, we identified three key performance metrics based on the standard for the daylighting design of buildings (
MOHURD of China, 2013a) and relevant previous studies (
Kim and Yi, 2019): building energy consumption, thermal comfort, and daylighting performance. The three optimization objectives―energy use intensity (EUI), predicted percentage dissatisfied (PPD), and useful daylight illuminance (UDI)―were selected specifically to address key performance priorities for residential buildings in cold-climate contexts. EUI directly corresponds to heating-dominated energy demands common in cold regions. PPD reflects occupant thermal comfort, crucial due to the substantial seasonal temperature fluctuations in these climates. UDI assesses daylight availability, critical for reducing artificial lighting energy consumption, particularly during shorter winter days.
Energy consumption evaluation indicator: Energy consumption significantly impacts the environment, so we used EUI as the primary evaluation metric. EUI represents the total annual delivered energy consumption per unit area of a building, measured in kilowatt-hours per square meter (kWh/m2). In this study, EUI refers to the final delivered energy for heating, cooling, lighting, and equipment operation, excluding upstream transmission losses and embodied energy. Energy loads were simulated using the Honeybee Energy module, based on EnergyPlus, under standardized occupancy and operational assumptions. Lower EUI values indicate higher energy efficiency, which is one goal of this study.
Thermal comfort evaluation indicator: Building energy consumption is directly impacted by the indoor thermal environment (
Zhu et al., 2020,
2022). Predicted mean vote (PMV) and PPD are internationally recognized performance indicators for evaluating indoor thermal environment quality (
Molake et al., 2023).In this study, PPD was selected as the thermal comfort indicator, with lower values indicating higher comfort. PPD values were calculated using the Ladybug PMV module based on hourly simulations throughout the entire year, and the reported values represent annual averages over 365 days. This annual averaging approach reflects the cumulative comfort performance under varying seasonal conditions and ensures consistency in evaluating form-based design impacts. The optimization objective was to minimize the annual mean PPD to improve overall thermal comfort in cold-climate residential environments.
Indoor daylighting performance evaluation indicator: Daylighting is crucial for energy consumption and occupant comfort (
Wang et al., 2021b). We employed UDI as the primary evaluation metric for daylighting performance, measuring the distribution of daylight illuminance within a specific range indoors to ensure adequate and comfortable lighting. In residential projects, UDI refers to the percentage of hours per year when natural light provides effective illuminance (300–2000 lx) (
John Mardaljevic et al., 2012). Daylighting simulations are generally conducted using Honeybee and Radiance software to ensure scientifically accurate results. Calculations are made between 8:00 a.m. and 4:00 p.m. throughout the year, with a working plane height of 0.9 m and a grid spacing of 3 m from the surrounding walls.
For this study, the three performance metrics established the MOO framework, aiming to offer a residential district design scheme with comprehensive performance optimization. This approach not only addresses single issues of energy efficiency or comfort, but also seeks to identify the optimal balance between energy efficiency, thermal comfort, and daylighting performance to achieve sustainable residential design objectives.
3.3 Simulation process
This study aimed to explore the MOO design of residential district spatial morphology in cold-climate regions of China. Primarily, we analyzed the comprehensive impact of building design variables on indoor natural lighting, thermal comfort, and energy consumption. The simulation process involved parametric modeling, building performance simulation, and MOO.
3.3.1 Generating parameter models
Parametric modeling was conducted using Grasshopper on the Rhino platform, known for its excellent visualization and extensibility, to enable effective simulation and analysis of residential district designs (
Sadeghipour Roudsari and Pak, 2013). By flexibly adjusting design parameters, we generated the initial residential models. Ladybug was used for the initial scene settings to ensure the model reflects real-world environmental conditions. After the appropriate simulation boundary conditions were defined, the parametric models were simulated using the Ladybug and Honeybee plugins within Rhino–Grasshopper, which serve as interfaces to core simulation engines, including EnergyPlus (via OpenStudio) for dynamic thermal and energy performance simulation, and Radiance/Daysim for annual daylight analysis. These tools were used to estimate energy use intensity (EUI), predicted percentage dissatisfied (PPD), and useful daylight illuminance (UDI) across varying design scenarios.
To ensure reliability and engineering applicability, all envelope parameters were set according to the Design Standard for Energy Efficiency of Residential Buildings in Severe Cold and Cold Regions (JGJ 26-2010) and relevant field data (
Gao et al., 2012;
MOHURD of China, 2021). As shown in Table 3, these include typical U-values for exterior walls, windows, roofs, and floors used in mainstream construction practice in northern Chinese cities. WWR, solar heat gain coefficients (SHGC), and air change rates were configured within code-compliant ranges, reflecting current engineering norms in the region. These settings ensure that our model represents a realistic baseline before morphological optimization.
Additionally, all non-optimized control parameters―such as indoor temperature setpoints, HVAC schedules, and internal load assumptions―were defined based on JGJ 26–2010, the Design Standard for Energy Efficiency of Residential Buildings in Shandong Province (DB37/5026-2022), and energy-use profiles from V.M. Barthelmes’ research (
Barthelmes et al., 2016;
MOHURD of China, 2021,
2022), as detailed in Table 4. Qingdao’s residential buildings predominantly use centralized heating systems. To isolate the impact of spatial morphology on operational energy use, heating demand was modeled as delivered thermal energy, assuming idealized system conditions without generation or distribution losses.
Total EUI was decomposed into heating, cooling, lighting, and equipment components in the baseline model: heating accounted for 74%, cooling 13%, lighting 4%, and equipment 9%. Equipment loads were kept constant across all scenarios, while lighting energy use varied according to daylight availability, influenced by building form and WWR.
The HVAC system was modeled using the Ideal Air Loads approach in EnergyPlus, with setpoints of 18 ºC for heating and 26 ºC for cooling, consistent with common practice in building performance simulations (
Feng et al., 2024;
Liu et al., 2023a).
Considering that residential districts usually contain hundreds of internal spaces and thermal zones, simulating a complete residential district model would be computationally intensive. To balance computational efficiency and accuracy, we modeled three representative floors (ground, middle, top) as single thermal zones per building. This strategy, validated in district-scale energy studies (
Sun et al., 2024;
Xiang et al., 2023), captures dominant morphology-driven performance trends while enabling tractable optimization across six typologies × 50 design variants × 20 generations = 6000+ scenarios, which would be computationally prohibitive under room-level resolution. We clarify that this simplification focuses on morphology-induced performance variations rather than intra-zone gradients.
3.3.2 Building performance simulation
At this stage, the building structure and design parameters were set, and Qingdao’s weather data files were imported from the EnergyPlus website to more accurately simulate the environmental conditions of the region (
U.S. Department of Energy, 2024). We then adjusted key design variables (e.g., OB, WWR, and Hb) to conduct the performance simulations using Honeybee and Ladybug. These components are widely used for building energy consumption, daylighting, and thermal comfort simulations and are reliable (
Ahmad et al., 2020;
Du et al., 2020;
Mardaljevic, 2016;
Reinhart and Fitz, 2006;
Ward et al., 2011). During the simulation process, we used version 1.7 of the Honeybee and Ladybug plugins, based on the Radiance and DAYSIM engines, to perform daylight and thermal comfort simulations (
Han et al., 2018). Energy consumption simulations were conducted using OpenStudio within Honeybee.
3.3.3 Multi-objective optimization
Genetic algorithms were applied for parameter optimization, utilizing the Wallacei plugin in Grasshopper to perform Pareto front analysis and identify optimal solutions in MOO. Wallacei is a MOO tool based on genetic algorithms and excels in handling complex problems requiring global search solutions, avoiding local optima, and identifying diverse and convergent Pareto optimal solution sets. Wallacei integrates multiple evolutionary algorithms, including NSGA-II, a common method for solving MOO problems, and thus achieves better overall performance by balancing different objectives (
Wang et al., 2005). Additionally, Wallacei provides users with a comprehensive set of evolutionary computation tools that can run evolutionary simulations, deeply analyze output results, select the best solutions using various methods (including K-means clustering), and extract and analyze detailed characteristics of specific solutions from all solutions (
Wallacei, 2022).
For this study, the MOO goals included minimizing the EUI of buildings, minimizing the PPD, and maximizing the UDI. These objectives reflect the key aspects of energy efficiency, indoor thermal environment, and daylight quality in residential district design. The optimization process considered algorithm elitism, mutation probability, mutation rate, and crossover rate. These settings align with GA-based residential thermal comfort optimization approaches, balancing computational efficiency and optimization effectiveness (
Baghoolizadeh et al., 2024). With these parameter choices, we aimed to balance the breadth and depth of the search space, ensuring the algorithm effectively explored the diversity of potential solutions; the specific data is shown in Table 5. The population size was set to 50, and the number of iterations was set to 20 generations. This configuration follows previous studies with similar variable dimensions and was further validated through pilot tests (
Molake et al., 2023;
Ziaee and Vakilinezhad, 2022). The initial population size (50) was validated via Latin Hypercube Sampling to achieve over 92% coverage of the discrete design space (780 combinations), ensuring robust sampling (see details in Appendix 2 Table A.4). Additionally, standard deviation graphs (see Appendix 2 Fig. A.7) showed that optimization trends stabilized by generation 10, confirming that 20 generations were sufficient for convergence. During optimization, design parameters such as OB; WWR for north, south, and west & east; and HB were strictly limited to the preset ranges to generate parameterized residential district models that meet actual engineering standards. The iterative optimization identified building configurations that provide optimal overall performance, and the correlations between these performance indicators and design variables were then analyzed.
This systematic method not only enhances the scientific rigor and accuracy of building design, but also provides practical design solutions and decision support for sustainably developing residential districts.
3.3.4 Model validation
To verify the realism of the simulation results and assess the applicability of the simplified modeling assumptions, we conducted a two-step validation process.
First, we benchmarked the baseline simulation outcomes against empirical data from a government-funded field study in Qingdao, covering 173,512 m2 of residential buildings across four districts. The simulated heating and total energy consumption differed from measured values by less than 5% (Fig. 4), supporting the validity of the Ladybug-Honeybee–EnergyPlus workflow for large-scale district applications in cold climates.
Second, to rigorously evaluate the simplified model’s accuracy, we constructed a refined reference model for a representative highrise building from Model 1. As shown in Fig. A.1 (in Appendix), this model subdivided each of the 16 floors into 15 thermal zones with realistic internal layouts and HVAC zoning. Annual performance comparisons (Table A.1) (in Appendix) revealed relative errors ≤12% for EUI, UDI, and PPD. Crucially, hourly statistical metrics (Table A.2) (in Appendix) for heating, cooling, lighting, and PPD were compared using regression analysis, RMSE, NRMSE, and Mean Absolute Percentage Error (MAPE), quantified the agreement.
- Heating: Coefficient of Determination (R2) = 0.972, Root Mean Square Error (RMSE) = 0.005 kWh/(m2·h), Normalized Root Mean Square Error (NRMSE) = 8.55%
- Cooling: R2 = 0.811, RMSE = 0.002 kWh/m2·h, NRMSE = 9.58%
- Lighting: R2 = 0.988, RMSE<0.001 kWh/m2·h, NRMSE = 2.80%
- PPD: R2 = 0.873, RMSE = 13.176, NRMSE = 0.12%
Regression scatter plots (Fig. A.2) visually confirm strong correlations for heating (R2 = 0.972), lighting (R2 = 0.991), and PPD (R2 = 0.962), while cooling (R2 = 0.811) shows moderate agreement. Error distributions (Fig. A.5) quantify deviations: >95% of heating errors fall within ±0.005 kWh/(m2·h), 92% of cooling errors within ±20%, and >90% of PPD errors within ±10%, with no systematic bias.
These results demonstrate that the simplified model reliably captures the directional trends and magnitude of performance predicted by the refined counterpart. High R2 values (e.g., 0.972 for heating, 0.988 for lighting), low RMSEs, and low NRMSEs (e.g., 0.12%–9.58%) confirm strong agreement between models. Visual comparisons in Fig. A.2 and error distribution histograms in Fig. A.5 further support this alignment. These findings affirm that the simplified model maintains adequate fidelity for early-stage multi-objective optimization, especially where trend identification outweighs absolute precision.
4 Results
4.1 Multi-objective optimization results
The MOO process was completed on a computer equipped with a Windows 10 operating system (AMD 77950X, 16 cores, 4.20 GHz, 32 GB RAM). Due to the complexity of the model and the variability of the objectives, each simulation took about 10 min, and producing one generation result took about 8.5 h. The entire optimization process lasted nearly 1000 h, with the final objective values stabilizing. A total of 6000 operations were executed during the optimization.
Figure 5 illustrates the spatial distribution of all feasible solutions generated during the optimization of the six residential models, with each point representing a spatial form of a solution and its unique genetic combination pattern. By recording the historical objective values of Pareto front solutions, we demonstrated the dynamic changes in the MOO process and presented these evolutions in detail through a three-dimensional scatter plot. The three axes in the figure represent UDI (%), EUI (kWh/m2/year), and PPD (%). Blue dots represent non-dominated Pareto front solutions, while grey dots indicate dominated solutions. To enhance interpretability, the original three-dimensional scatter plot has been decomposed into three pairwise 2D projections―UDI vs. PPD, EUI vs. UDI, and PPD vs. EUI―-shown alongside the 3D plot. Red dashed arrows indicate the direction of performance improvement for each objective pair; better performance is indicated by a higher UDI and a lower EUI or PPD.
To quantify and compare the performance of the optimal solutions for each model, we set initial values for each model based on commonly used market parameters. These initial values helped facilitate clear comparisons of the effects of the optimization. Referring to commonly used parameters in the current residential design market, we set the initial values for energy consumption (EUI), daylight (UDI), and thermal comfort (PPD) under the following conditions: an HB of 3 m, 30º east of south OB, a north WWR of 0.3, a south WWR of 0.4, and an east-west WWR of 0.3. These values served as comparative references for the optimization, helping with the evaluation of any improvement in model performance.
Initial values, Individual Optimal Values (i.e., best performance for each individual objective), and Overall Optimal Values (i.e., the best balanced Pareto solution chosen via average fitness ranks) are shown in Table 6 along with the associated Improvement (%) for each model. These data support both horizontal comparisons across models and vertical analyses within each objective. Different indicators have different optimization goals. For EUI and PPD, a lower value denotes better performance (i.e., less energy consumption and increased thermal comfort), while a greater value is preferable for UDI (i.e., increased daylight availability). Therefore, an improvement over the baseline is shown by a positive Improvement (%), whereas a decline in performance is indicated by a negative value.
From the perspective of the best optimal solution, Model 2 showed positive optimization in all three targets (EUI, UDI, and PPD), indicating an overall improvement in performance. Among the three objectives, UDI exhibited the greatest improvement, with a maximum percentage increase of 7.92%, followed by EUI, with a maximum percentage reduction of 6.14%, and PPD, with a maximum percentage reduction of 4.84%.
In terms of EUI, all six models exhibited consistent downward trends compared to the initial configurations, with relative reductions ranging from 3.94% to 6.14%. Interestingly, Model 3 showed the most substantial decline. Similarly, UDI values improved across all models, particularly for Model 6 (up to 7.92%) and Model 4 (7.48%). For thermal comfort, all models demonstrated reduced annual PPD values, with the greatest relative decreases observed in Model 4 (4.84%) and Model 2 (4.35%). Although the absolute magnitudes of improvement are moderate, they reflect robust optimization tendencies under early-stage assumptions.
Overall, these MOO models showed significant improvements in UDI, EUI, and PPD and thus demonstrated the effectiveness of MOO in balancing different performance indicators. Model 4, the Slab-Mixed Height type, especially showed significant improvements in EUI, UDI, and PPD. These improvements indicate that in practical applications, the MOO method can effectively enhance the building performance of such spatial morphology residential districts.
4.2 Analysis of the distribution and trends of multi-objective optimization solutions
The optimization outcomes for the six spatial models across the three performance targets―EUI, UDI, and PPD―are evaluated on two levels in this section. To evaluate performance stability and objective-wise advantages, Section 4.2.1 uses box plots to compare the whole distribution of Pareto-optimal and dominated solutions. The performance trends of the top-ranked Pareto solutions are examined in Section 4.2.2, which also highlights form-specific optimization priorities and tradeoffs under high-performance conditions.
4.2.1 Performance comparison between Pareto and dominated solutions
A detailed comparison was conducted between Pareto-optimal and dominated solution sets for all six spatial models, highlighting differences in EUI, UDI, and PPD performance. Boxplots show the performance distributions of the Pareto-optimal solutions (blue) in relation to dominated solutions (grey) for each of the six spatial models, as shown in Fig. 6.
In general, the Pareto-optimal solutions for EUI (Fig. 6(a)) showed smaller interquartile ranges and much lower median values than their dominated counterparts, suggesting improved solution stability and energy performance. Specifically, a number of models displayed Pareto fronts that were highly concentrated and had little dispersion, indicating that the optimization technique was able to filter out superior options.
For UDI (Fig. 6(b)), similar trends were observed, with Pareto-optimal solutions achieving higher median values and narrower ranges. These outcomes suggest that morphological optimization not only improves daylighting performance but also enhances the uniformity and predictability of simulation results, which is critical for robust design decision-making.
The Pareto-optimal solutions continuously outperformed dominant ones in terms of PPD (Fig. 6(c)), exhibiting lower discomfort rates and less variability. This outcome demonstrates how well the MOO system can provide options between illumination, comfort, and energy goals.
Overall, these comparisons demonstrate that Pareto-optimal solutions are superior in terms of statistical stability and dependability in addition to mean performance. Through a formal variability analysis, the following section (4.3.2) delves deeper into the ways in which various building form typologies contribute to these performance disparities.
4.2.2 Performance trends and trade-offs among Pareto-optimal solutions across spatial models
To extract representative solutions for performance interpretation, we applied a distribution-based filtering and ranking strategy to the Pareto fronts of each spatial model. Figure 5 illustrates the spatial distribution of Pareto-optimal solutions across the three objectives―EUI, UDI, and PPD. Distinct directional clusters were observed: some regions were dominated by energy-efficient solutions (low EUI), others by daylight-optimized configurations (high UDI), and some by thermally comfortable options (low PPD). These trade-off zones also corresponded to certain morphological patterns―for instance, slab-type forms (e. g., Model 3 and Model 4) frequently occupied low-EUI regions, while point-slab hybrids (e.g., Model 5) appeared more often in comfort-optimized clusters. These observations reveal interpretable performance tendencies linked to design typologies, providing intuitive guidance for early-stage form selection.
To capture these tendencies in a concise yet representative manner, we adopted the Average Fitness Rank method provided by Wallacei to identify Pareto solutions with balanced performance across all three objectives. For each model, the top three ranked solutions were extracted and averaged to form a representative configuration. This strategy avoids reliance on outlier solutions and better reflects the typical optimized behavior of each spatial typology.
Each representative solution was then mapped back to its corresponding parametric model, enabling a direct link between numerical performance and morphological configuration. This mapping facilitates design interpretation by allowing architects to evaluate how specific spatial forms―such as slab-dominant or point-slab hybrid layouts―behave under integrated performance criteria.
To facilitate visual comparison across objectives, all performance metrics were normalized to a [0,1] scale, where higher values indicate better outcomes (i.e., lower PPD and EUI, higher UDI). The results are displayed in Fig. 7, and the corresponding variable settings are listed in Table 7. This comparative visualization high-lights the relative strengths and trade-offs of each model.
- Model 1 (Point-Highrise) exhibited the weakest overall performance, with the highest EUI (worst energy efficiency) and lowest UDI (worst daylighting performance) among all models. Although its thermal comfort is slightly better than average, this model is not recommended for projects prioritizing energy efficiency or daylight access.
- Model 2 (Point-Mixed Height) achieved the highest UDI performance, indicating excellent daylight conditions. However, its thermal comfort and energy efficiency were less favorable than those of other models.
- Model 3 (Slab-Highrise) demonstrated strong energy efficiency and good daylighting performance. Its thermal comfort was moderately good, making it a suitable option for energy-prioritized projects with balanced comfort needs.
- Model 4 (Slab-Mixed Height) had the lowest EUI (best energy performance), but exhibited higher PPD values (thermal comfort is limited). This model may be favored in energy-focused projects but requires supplemental comfort strategies.
- Model 5 (Point-Slab Hybrid-Highrise) demonstrated the best thermal comfort performance among all models, while achieving moderate results in both EUI and UDI. This makes it a suitable option for scenarios prioritizing occupant comfort without significantly compromising energy efficiency or daylighting.
- Model 6 (Point-Slab Hybrid-Mixed Height) showed the weakest thermal comfort performance with the highest PPD among all models. Its UDI and EUI values were moderate, indicating no outstanding strength in any particular objective. Therefore, this model is less suitable for designs aiming for integrated comfort and performance.
These findings show that the spatial arrangement of each model affects how it reacts to multi-objective tradeoffs. The most well-balanced approach is Model 3 (Slab-Highrise), which achieves stable thermal comfort, great daylight availability, and outstanding energy efficiency. Model 2 (Point-Mixed Height) is better suited for daylight-focused situations because it has a slightly higher PPD and EUI but excels in UDI. With the lowest EUI, Model 4 (Slab-Mixed Height) is the best option for energy-conscious design schemes. On the other hand, Model 5 (Point-Slab Hybrid-Highrise) has the lowest PPD values, which makes it ideal for applications where comfort is a top priority. Although the performance gaps among top solutions are not always dramatic, they reflect consistent and interpretable optimization tendencies.
It is important to note that these findings are based on a limited number of high-performing Pareto solutions and are intended to inform design selection under constrained, high-performance scenarios. Rather than being statistically generalizable, they represent the inclinations of each model under idealized conditions. Section 4.3.2 further addresses this by providing a non-parametric statistical analysis of the entire solution dataset, providing more reliable and transferable findings for real-world design decision-making.
4.3 Correlation and variability analysis of design variables
Statistical analysis in this study was conducted using SPSS 26.0, divided into two main parts. The first part was correlation analysis using the Spearman rank correlation coefficient to identify the relationships between numerical design variables (including OB, WWR, and HB) and optimization objectives (PPD, UDI, and EUI). Overall, this preliminary screening helped us understand the associations between these variables. The second part was variability analysis using the Kruskal–Wallis test to evaluate the differences in optimization objectives among categorical design variables such as plan layout types (Point Layout, Slab Layout, Point-Slab Hybrid Layout) and vertical distribution types (Highrise Only, Mixed Height). For further pairwise comparisons of categorical variables, we used Dunn’s test. These analytical methods helped us understand the relationships between specific spatial morphology and optimization performance via exploring the underlying logic of the performance characteristics, to gain a more comprehensive understanding of the impact of design variables on optimization objectives. The significance level was set at 0.05.
4.3.1 Correlation analysis of numerical design variables
We recorded all spatial morphology factors and performance simulation results generated in each iteration during the MOO process. To quantitatively study the relationship between spatial morphology factors and EUI, PPD, and UDI, a heat map was drawn, and the Spearman correlation analysis was conducted using SPSS software. Figure 8 illustrates the correlation between design variables (including OB, WWR-N, WWR-S, WWR-WE, HB) and target values (PPD, UDI, EUI). The color depth and block size represent the size of the correlation coefficient, Spearman’s rho is marked on the color block, and significance is indicated with asterisks to quickly show the correlation between dependent and independent variables. Below, the detailed analysis results are explained.
For EUI, no significant correlation existed between EUI and OB, WWR-N, and WWR-WE (
p < 0.05), indicating that these variables minimally impact EUI. Consistent with integrated sensitivity frameworks (
Li et al., 2025), our analysis confirms WWR-S as the dominant variable for EUI (
r = – 0:61), highlighting its critical role in cold-climate energy savings. EUI was significantly positively correlated with HB (
r = 0:28;
p < 0:05), indicating a weak correlation where increasing HB leads to increased EUI, possibly due to the larger space requiring more heating and cooling. EUI was significantly negatively correlated with WWR-S (
r = – 0:61;
p < 0:05); this finding indicates a strong correlation where EUI is reduced when WWR-S increases. Among the factors influencing EUI, the WWR-S (
r = – 0:61) was the most significant, suggesting that optimizing the WWR-S is crucial for reducing EUI.
For UDI, no significant correlation existed between UDI and OB (p > 0:05), indicating that OB has little effect on natural lighting. However, UDI was significantly positively correlated with HB (r = 0:40; p < 0:05), indicating a moderate correlation where increasing HB enhances UDI. UDI was also significantly positively correlated with the WWR-N (r = 0:18; p < 0:05), but r < 0:3 indicates a weak correlation. In contrast, UDI was significantly negatively correlated with the WWR-S (r = – 0:87; p < 0:05) and the WWR-WE (r = – 0:32; p < 0:05), indicating a strong negative correlation for the WWR-S and a moderate negative correlation for the WWR-WE. These findings suggest that increasing WWR-S and WWR-WE may reduce UDI due to issues such as glare and excessive light. Overall, the WWR-S (r = – 0:87), HB (r = 0:40), and WWR-WE (r = – 0:32) were the most significant factors for UDI, and adjusting these three design variables appears crucial for optimizing UDI.
For PPD, the significant correlations with design variables mirrored those observed for UDI. PPD exhibited no significant correlation with OB (p > 0:05), implying that OB has minimal impact on PPD. Conversely, PPD was significantly positively correlated with HB (r = 0:45; p < 0:05), indicating a moderate relationship where an increase in HB leads to higher PPD. This increase may be attributed to greater air stratification, resulting in discomfort in taller spaces. Additionally, PPD had a weak positive correlation with WWR-N (r = 0:22; p < 0:05) and strong and moderate negative correlations with WWR-S (r = – 0:85; p < 0:05) and WWR-WE (r = – 0:41; p < 0:05), respectively. Thus, increasing WWR-N diminished comfort, whereas enhancing WWR-S and WWR-WE enhanced comfort. Comparing correlation coefficients shows that the WWR-S (r = – 0:85), HB (r = 0:45), and WWR-WE (r = – 0:41) are the most significant factors, and optimizing these three indicators is key to improving comfort.
These findings have particularly important design implications for cold-climate residential projects. The negative correlation between WWR-S and EUI (r = – 0:61) indicates that increasing the window-to-wall ratio on the south facade can effectively reduce heating energy demand through enhanced passive solar gain during winter months. Given that south-facing facades in northern latitudes receive the most stable solar radiation, this correlation reaffirms traditional passive design strategies, such as prioritizing solar exposure on southern elevations.
However, while a higher WWR-S can reduce EUI, excessive glazing may also lead to thermal discomfort, glare, and increased nighttime heat loss. Therefore, in practice, architects should consider optimizing WWR-S within a reasonable range (e.g., 0.30–0.45 in this study) and combine it with high-performance glazing, thermal breaks, and seasonally adaptive shading devices such as overhangs or louvers. These strategies can help maximize winter solar gain while avoiding overheating or inefficiency in transition seasons. This data-driven insight supports more nuanced south facade design decisions in cold-climate residential architecture.
4.3.2 Statistical evaluation of building form variability
Non-parametric tests were performed on the set of Pareto-optimal solutions to assess the relationship between plan layout types, vertical distribution types, and their combined forms concerning the three target variables: EUI, UDI, and PPD. It was carried out for insight into the performance impact of various spatial configurations. The Kruskal–Wallis tests were used since the categorical variables did not satisfy the normality requirements. Three plan layout types were examined ―Point Layout, Slab Layout, and Point-Slab Hybrid Layout―and two vertical distribution types―Highrise Only and Mixed Height. These were further combined into six representative building forms: Model 1 through Model 6 (see Section 3.2.1), enabling a more comprehensive understanding of how spatial form influences performance.
4.3.2.1 Impact of plan layout types
The test results (Table 8) revealed that plan layout types significantly affected EUI and PPD (p < 0:01), but not UDI. In terms of EUI, Slab Layout (median 122.755) performed best, while Point Layout (median 131.080) and Point-Slab Hybrid Layout (median 127.530) performed poorly (p < 0:01). This implies that slab types, maybe as a result of their compact design and greater south-facing facades, are more efficient in lowering energy demand. The Slab Layout exhibited marginally lower PPD values (median = 48.85) compared to both the Point Layout (48.96) and Point-Slab Hybrid Layout (49.25), suggesting a measurable, though limited, improvement in thermal comfort performance. This subtle variation highlights the potential influence of spatial layout strategies on occupant comfort optimization in built environments.
4.3.2.2 Impact of vertical distribution types
Analysis of vertical distributions (Table 9) indicated significant differences in UDI and PPD, but not in EUI. Mixed Height (median 87.61) had significantly higher UDI values than Highrise Only (median 86.23) (p < 0.05). Mixed Height may provide better natural lighting environments, especially with significant improvements in lighting conditions on lower floors. However, Highrise Only performed better in terms of thermal comfort, with a lower PPD value (median 48.52 vs. 49.42, p < 0.01), which may be attributed to enhanced natural ventilation at higher elevations and reduced thermal stagnation.
4.3.2.3 Combined impact of building forms
By combining plan layout types and vertical distribution types, the six composite building forms were analyzed (Table 10, Fig. 9). The results showed significant differences across all three performance indicators (p < 0.05). It is noted that Dunn’s post-hoc tests revealed distinct performance tiers denoted by grouping letters (a, b, c) in Table 10, with critical pairwise differences marked in Fig. 9. Post-hoc Dunn’s tests with Bonferroni adjustment quantified pairwise differences, with full results in Tables A5-A7 (Appendix 3).
In terms of EUI, Model 3 (Slab-Highrise) and Model 4 (Slab-Mixed Height) performed best (Group a; medians 122.58 and 122.93 kWh/m2), while Model 2 (Point-Mixed Height) had the highest EUI (Group c; 131.29 kWh/m2). The advantage of slab layouts in energy efficiency was statistically confirmed (p < 0.001 vs. point forms).
Regarding UDI performance, Model 2 (Point-Mixed Height) achieved the highest UDI (Group c; median 89.73%), significantly outperforming most models (p ≤ 0:003) except Model 3 (p = 0:077). Model 1 (Point-Highrise) showed the lowest UDI (Group a; 83.28%), while the UDI values of the other four types did not show significant differences and thus further validated the advantage of the Point-Mixed Height in daylight utilization, as indicated by the non-parametric tests of vertical distribution.
For PPD, Highrise Only configurations―Model 1, 3, and 5 (Group a)―consistently provided better thermal comfort (medians 48.50–48.55%), significantly surpassing mixed-height forms as shown in Fig. 9(c) (p < 0.001). Model 6 (Hybrid-Mixed) exhibited the highest discomfort (Group c; 49.92%, p < 0.001 vs. all).
The statistical results confirm that both plan layout and vertical distribution significantly affect energy, lighting, and comfort performance. Specifically.
- Slab-Highrise forms are best suited for minimizing energy use;
- Point-Mixed Height forms excel in daylighting performance;
- Highrise Only schemes offer better thermal comfort.
Designers can use the grouped performance summaries in Table 10 to guide form selection. The optimization-based conclusions in Section 4.2.2 are in agreement with the statistical results of Section 4.3.2. Both analyses identify Model 3 (Slab-Highrise) as the most well-rounded performer, demonstrating strong median values across all three objectives―relatively low EUI and PPD, and high UDI―making it the most statistically robust candidate for balanced performance.
These results reflect the complementary strengths of optimization-based and population-based evaluations. The former informs solution-oriented early design exploration, while the latter provides generalizable evidence for reliable decision-making. Together, they provide a comprehensive framework for selecting high-performance spatial configurations in cold-climate residential planning.
5 Discussion
5.1 Main findings
This research examined cold-climate cities in China, utilizing MOO and quantitative analysis to explore the effects of various residential spatial morphologies on building energy consumption, indoor thermal comfort, and useful daylight illuminance. For that exploration, we introduced a MOO workflow based on Pareto optimal solutions, employing Rhino Grasshopper for baseline model construction, along with Wallacei tools. We used that methodology to optimize the residential orientation, height of building, and window-to-wall ratio, aiming for minimal EUI, optimal PPD, and maximum UDI. Through Spearman correlation analysis and non-parametric tests, we then analyzed the relation-ships between building design variables (e.g., orientation, WWR, and height) and performance indicators, as well as the performance differences across different building form types.
Considering the results of the initial setups, there were substantial improvements in performance for each of the three objectives. Specifically, EUI was reduced by 3.94%– 6.14%, UDI increased by up to 7.92%, and PPD was reduced by up to 4.84%. Compared with similar residential optimization studies, our method demonstrated competitive or superior performance. A recent urban-scale optimization framework targeting residential layout performance reported a 1.5% reduction in energy use, alongside large improvements in photovoltaic potential and daylight hours (
Liu et al., 2023a). Zhu et al. conducted a multi-objective optimization of rural residential typologies in northern China, achieving a maximum EUI reduction of 4.85% (
Zhu et al., 2020). Similarly, in Jinan―another cold-climate city―optimization of residential layouts reduced heating energy consumption by approximately 5.9% (
Deng et al., 2021). Furthermore, a morphology-based optimization of 96 residential clusters in Hangzhou achieved an average EUI reduction of 7.73% (
Feng et al., 2024), aligning closely with our results. Given that heating loads represented 74% of total EUI and were the primary target of form-based improvements, the observed reductions are attributable to enhanced passive performance through spatial reconfiguration.
Compared with previous studies, our UDI improvement (7.92%) may appear modest. However, this is largely due to the high baseline value of 82.82% associated with the high-rise forms studied. For example,
Zhu et al. (2020) investigated three types of low-rise courtyard houses and reported sDA improvements ranging from 21.33% to 26.45%, based on much lower initial values (e.g., 34.48%–76.23%) (
Zhu et al., 2020). Their larger improvement range reflects the greater potential for daylight enhancement in low-rise rural dwellings. In contrast, our findings demonstrate that even within highly lit, high-density urban conditions, targeted morphological optimization can yield measurable gains in daylight autonomy. Another MOO framework at the urban planning stage achieved a 50% increase in daylight hours and a 52.7% improvement in photovoltaic potential (
Liu et al., 2023a). Although these values are considerable, they often rely on idealized geometries or low-density scenarios. Conversely, given practical restrictions, a more thorough optimization combining envelope properties and control variables only produced a 1.7% UDI increase (
Wu et al., 2024). Wu et al. conducted their study in the hot-summer cold-winter climate zone, focusing on the simulation and optimization of a single residential building. The optimization variables included building orientation; thermal transmittance of exterior walls, windows, floor slabs, and roofs; window solar heat gain coefficient; air changes per hour; and window-to-wall ratios for four orientations. As the study did not involve morphological layout modifications, the potential for improving daylight performance was inherently limited, likely contributing to the modest UDI improvement reported. Our UDI improvement of 7.92%, though moderate compared to morphology-driven benchmarks, substantially outperformed envelope-based approaches under comparable conditions. This increase was primarily driven by enhanced daylight penetration through mixed-height distributions and south-facing window optimization. These findings underscore the high sensitivity of daylight performance to spatial configuration and confirm the value of morphological levers in dense urban design contexts.
Regarding thermal comfort, the maximum PPD reduction achieved was 4.84%, with consistent improvement trends across most models. This improvement is small, but it is comparable to or better than previous research. In northern China, for instance, the optimization of rural courtyard patterns revealed a 3.05% PPD improvement (
Zhu et al., 2020). Another study that focused on ultra-low energy buildings in Changsha revealed reductions ranging from 3.52% to 11.09%, depending on user behavior and retrofit technique (
Xiang et al., 2023). An envelope-based optimization study also obtained a 10.9% reduction in discomfort rating (
Wu et al., 2024;
Xiang et al., 2023). In contrast, our optimization focused solely on early-stage form variables―such as height, orientation, and WWR―without altering materials or HVAC systems. This constraint naturally limits potential gains, yet the results still demonstrate that morphological adjustments alone can yield up to a 5% reduction in annual discomfort, especially in high-density contexts where design flexibility is constrained during early planning.
Even though the PPD improvement in absolute terms seems little, it is a significant yearly average that was obtained from many simulations conducted over a full year. The observed optimization trends were constant among models, despite the fact that early-stage simulations invariably involve some simplifications (such as thermal zoning or generalized internal loads). Additionally, the directionality and trade-offs of improvement, which are confirmed by non-parametric statistics and Pareto front analysis, continue to be strong and interpretable, confirming the validity of the results for early planning guidance.
Figure 10 shows daily PPD fluctuations for the six models’ best solutions throughout the duration of the year to more accurately evaluate thermal comfort under various seasonal conditions. For clarity, the hot season (June–August) and cold season (November–March) are shaded. PPD levels exhibit strong seasonal patterns, with consistently high values (60%–80%) during the cold season (November–March), peaking near 90% as shown in monthly trends (Fig. A.3). This elevation is partially attributed to simplified thermal zoning (one zone per floor), which blends private and shared spaces, amplifying discomfort metrics. Extreme-day analyses (Fig. A.4) further reveal that PPD remains below 30% across all models during the hot season, primarily due to Qingdao’s mild summers and enhanced ventilation achieved through optimized morphological configurations. Strategies such as increased WWR on the south façade and mixed-height layouts contribute to improved summer comfort by promoting solar control and airflow. However, their effectiveness is limited in winter due to persistent heating demand. These findings affirm that while model simplifications accentuate absolute PPD values, morphological optimization remains valuable for early-stage design―particularly in enhancing comfort during summer months―though supplemental heating measures are necessary to address winter limitations.
Our optimized residential models achieved EUIs between 118.58 and 128.30 kWh/m
2/year, notably lower than previously reported energy benchmarks. According to the Annual Development Research Report on Building Energy Efficiency in China (
Building Energy Efficiency Research Center of Tsinghua University, 2018–2022), typical energy consumption for existing residential buildings in northern China ranged from 80 to 150 kWh/m
2/year, with heating accounting for approximately 60%–70%. Further, measured data from retrofit projects in northern regions, such as in Shenyang (133 kWh/m
2/year) (
MOHURD of China, 2009), confirmed the effectiveness of our optimization strategies in achieving meaningful energy savings compared to common regional benchmarks.
The correlation analyses provided further insight into influential design parameters. Increasing the WWR-S significantly decreased EUI and improved thermal comfort, although excessive window ratios could adversely impact daylight suitability. These results highlight the delicate balance needed between window size, daylight autonomy, and energy efficiency and are in alignment with earlier research by
Kamal et al. (2021) and
Sorooshnia et al. (2022). Moreover, increased floor height positively impacted daylight levels but reduced indoor thermal comfort, echoing the insights from Elgheriani and Cody’s findings on highrise residential buildings (2019).
A notable divergence in results pertains to building orientation. Although orientation was found to significantly affect energy performance in Jinan (with optimal orientations around 15º from south) (
Deng et al., 2021), our analysis showed no statistically significant orientation effect on EUI, UDI, or PPD in Qingdao’s high-density residential contexts. This discrepancy may be attributed to differing urban densities and building typologies affecting solar access and wind-driven ventilation. While several studies in cold-climate regions (e.g., Jinan) found that orientations within ±15º of due south significantly reduced heating demand due to better solar exposure and prevailing wind alignment (
Deng et al., 2021;
Toparlar et al., 2015;
Wang et al., 2021a), these effects may diminish in high-density urban contexts like Qingdao, where inter-building shading and reduced ventilation potential limit the influence of orientation. This suggests that the impact of orientation is context-dependent and highlights the need for region- and density-specific design strategies.
The Pareto optimal solutions clearly illustrated how plan layout and vertical distribution significantly influence building performance, highlighting the inherent trade-offs among energy use, daylighting, and thermal comfort. In particular,
Slab Layout outperformed
Point Layout and
Point-Slab Hybrid Layout configurations, exhibiting the lowest EUI and improved thermal comfort. This benefit mainly results from the slab layout’s bigger south-facing facade, which minimizes heat loss because of the lower facade exposure per unit area and encourages passive solar gain. Furthermore, well-planned peripheral layouts can lower energy usage and improve thermal comfort, high-lighting the crucial role that morphology plays in early-stage design (
Li et al., 2024).
Point Layout, on the other hand, used more energy and was less comfortable. This was mostly because of its larger surface-to-volume ratio and increased exposure to cold winter winds, which caused more heat loss. Our findings indicated that
Mixed Height provided better daylight conditions than uniform
Highrise Only forms, as height variation effectively disrupted continuous shading, enabling increased direct sunlight access at lower floors (
Yazdandoust, 2024).
Remarkably, when it related to thermal comfort,
Highrise Only layouts outperformed
Mixed Height setups. This is because, at higher elevations, they are more exposed to wind, which enhances convection cooling and reduces localized heat retention during hot seasons, especially in densely populated urban areas (
Zhang et al., 2023).
The comparative results show that no one type prevails across all parameters when viewed from a multi-objective perspective. Rather, distinct tradeoffs are shown by various spatial arrangements, which need to guide further design strategies as discussed in Section 5.2.
These suggestions are in alignment with previous research on urban morphology and building performance in addition to our simulation results. Optimized residence layouts have been repeatedly demonstrated in prior studies to effectively lower energy demand, improve ventilation, and improve microclimatic conditions (
Feng et al., 2024;
Yan et al., 2023b). For example, it has been proven that greater height variability improves heat dissipation and lowers summer outdoor temperatures, which decreases the need for cooling energy (
Perini and Magliocco, 2014;
Skarbit et al., 2017). On the other hand, it has been discovered that overly deep urban canyons restrict ventilation and worsen thermal discomfort (
Krishan, 2001). High-density and compact building configurations have been recognized as universally beneficial for energy efficiency across various climatic conditions (
Trepci et al., 2021;
Vartholomaios, 2017). This underscores the dominant tendency towards dense urban residential development, aimed at meeting both economic feasibility and sustainability goals. Overall, our findings indicate that educated morphological decisions, led by multi-objective assessments, serve as key mechanisms for establishing sustainable, comfortable, and energy-efficient residential settings in cold-climate urban contexts.
Moreover, both the building forms and envelope configurations used in our simulations were carefully calibrated to match actual practice. The six form types were extracted from a survey of 660 residential districts in Qingdao. These considerations ensure that our findings are not only theoretically robust but also practically applicable to real-world residential development scenarios in cold-climate regions.
5.2 Morphological design strategies
Based on the identified performance characteristics and underlying mechanisms from the above analysis, this study proposes several clear morphological strategies to guide residential design in cold-climate urban contexts.
First, optimize WWR. The results showed that increasing WWR-S significantly reduces energy use and enhances thermal comfort due to improved passive solar gain. However, excessive increases in WWR-WE, while beneficial for daylight access, may negatively impact comfort and energy efficiency due to excessive solar radiation and heat loss. Hence, a balanced and orientation-sensitive window design approach should be adopted, corroborating previous findings on fenestration design impacts on building performance (
Kamal et al., 2021;
Sorooshnia et al., 2022).
Second, carefully consider vertical distribution. Mixed Height distributions demonstrated better daylighting performance by effectively reducing mutual shading among buildings, enhancing direct daylight penetration, especially to lower floors. Conversely, uniform
Highrise Only forms provided superior thermal comfort due to improved convective airflow and reduced urban heat retention in upper floors. This suggests that vertical height differentiation strategies should be prioritized if daylight optimization is the primary goal, while consistent high-rise strategies have a more direct positive impact on thermal comfort (
Yazdandoust, 2024;
Zhang et al., 2023).
Third, select an appropriate plan layout type. In comparison to
Point Layout or
Point-Slab Hybrid Layout,
Slab Layout continuously showed better energy and comfort performance due to its bigger south-facing surface areas and compact building geometries. The
Point Layout, on the other hand, showed worse energy efficiency and thermal comfort because of increased facade exposure, but greatly enhanced daylight conditions at higher elevations since there were fewer inter-building impediments. Hence, the
Slab Layout is recommended when energy and thermal comfort are prioritized (
Li et al., 2024).
Finally, comprehensively integrate vertical and plan layout forms. For balanced performance across energy, daylighting, and comfort, the Slab-Highrise configuration consistently stands out across both Pareto-based and statistical evaluations. It exhibits low EUI, stable PPD, and high UDI, making it a robust candidate for projects seeking comprehensive performance.
These morphological design strategies, derived from rigorous multi-objective analysis and corroborated by existing literature, can help practitioners effectively address competing performance requirements, contributing significantly to the development of sustainable, comfortable, and energy-efficient residential communities in cold climates.
5.3 Design guidelines for target-oriented planning
Building upon the morphological performance trends identified in Section 5.2, this section distills practical design strategies tailored to three typical planning priorities―energy efficiency (EUI), daylight sufficiency (UDI), and thermal comfort (PPD)―as well as an integrated tradeoff solution. Each recommendation is grounded in Pareto-optimal solutions and statistically validated outcomes (Table 10, Section 4.3), offering actionable guidance for planners and architects.
Design strategy 1: Balanced performance across all metrics (EUI, UDI, PPD).
● Recommended form: Slab-Highrise.
● Key parameters: WWR-S = 0.45–0.50; WWR-WE ≤ 0.30; Highrise only or slightly mixed height.
● Rationale: Dominates central region of Pareto front, offering robust compromise performance.
● Note: Best suited for projects with multiple conflicting targets and regulatory constraints.
Design strategy 2: Maximize energy efficiency (minimize EUI).
● Recommended form: Slab-Highrise.
● Key parameters: WWR-S = 0.45–0.50; WWR-WE ≤ 0.25; Highrise only (height = 54–80 m).
● Rationale: Minimizes surface-to-volume ratio and maximizes passive solar gain (EUI median 122.58 kWh/m2).
● Note: Optimal under compact site conditions with low shading risk.
Design strategy 3: Maximize daylighting (maximize UDI).
● Recommended form: Mixed-Height + Point Layout.
● Key parameters: Mixed height ratio ≈1:1 (high-rise + mid-rise); Spacing ≥1.2 × building height; Moderate WWR-WE = 0.30–0.35.
● Rationale: Breaks shading continuity, improves daylight access to lower floors (UDI median 89.73%).
● Note: May slightly compromise thermal comfort; mitigation via shading devices and adaptive ventilation.
Design strategy 4: Maximize thermal comfort (minimize PPD).
● Recommended form: Slab Layout + Highrise Only.
● Key parameters: WWR-WE ≤ 0.25; Orientation: 0º–10º (east of south); Building height >54 m with slab configuration.
● Rationale: Promotes stable airflow and minimizes solar asymmetry (PPD median 43.25%).
● Note: Avoid excessive glazing on east/west facades.
These goal-oriented strategies translate complex optimization outcomes into actionable design principles, offering clear guidance for urban planners and architects navigating trade-offs in cold-climate residential projects. By aligning form typologies and parameter settings with specific performance priorities, they support evidence-based decision-making in the early design stage.
5.4 Limitations
Despite the positive results obtained through multi-objective optimization (MOO), this study has several limitations that need to be addressed in future research.
Selection of optimization objectives: The current study focuses on three conflicting objectives: energy use intensity (EUI), predicted percentage of dissatisfied (PPD), and useful daylight illuminance (UDI). These objectives offer a comprehensive evaluation of building performance; however, there are additional objectives such as glare control, ventilation, and indoor air quality that were not considered in this study. Including these factors could provide a more holistic evaluation of building performance.
Furthermore, the Pareto optimal front generated in this study is limited to the three objectives mentioned. If a designer or engineer wishes to optimize for additional objectives, such as acoustic comfort or economic cost, the current optimization approach may face challenges. While the NSGA-II algorithm used in this study performs well for three objectives, it may not be as efficient when handling more than three conflicting objectives.
Computational complexity: The current optimization process takes approximately 1000 h due to the complexity of the simulations. This presents a major limitation in real-world applications where faster feedback is required. Future research should explore ways to reduce the computational burden, such as using surrogate models or artificial neural networks (ANNs) to accelerate the optimization process while maintaining accuracy.
Model simplification trade-offs: The use of a simplified thermal zoning method―modeling three representative floors and whole-floor thermal zones―greatly reduced simulation time (~20 × faster than refined models), enabling large-scale MOO. Additionally, the simplified thermal zoning approach inherently neglects vertical temperature stratification, potentially influencing detailed thermal comfort predictions. Future studies may incorporate refined zoning methods or coupling with CFD modeling to explicitly capture height-dependent temperature gradients for improved accuracy at later design stages. For daylighting simulations, the simplified model employed fewer sensor points than the refined counterpart, but UDI distributions remained statistically comparable (Fig. A.6), justifying its use for early-stage comparative studies where relative performance matters more than absolute values. While this simplification inherently introduces abstraction and leads to minor deviations in absolute performance metrics (relative errors ≤12%), hourly statistical validation confirms robust directional accuracy (R2 ≥ 0.81, NRMSE ≤9.58%) across key indicators (Table A.2). This makes the approach suitable for early-stage planning where trend identification is prioritized over precision. As part of this validation, one typical form (Model 1) was randomly selected for detailed comparison. Due to resource constraints, fully refining all six forms was not feasible, and the goal was solely to verify zoning simplification rather than morphology-specific outcomes. To further isolate the morphological impact, only three early-stage design variables were included. Other influential parameters―such as facade materials, shading strategies, and floor area ratio―were intentionally fixed based on typical values and regulatory references, to control for confounding effects and ensure result robustness. This allowed clearer attribution of performance changes to form alone, but may limit comprehensiveness. Future research could integrate envelope-level and system-level variables into multi-scale optimization frameworks for a more holistic assessment.
Practical implementation constraints: In addition, project-specific constraints may limit the applicability of the proposed morphological strategies in real-world scenarios. Although all optimized design variables in this study―such as floor height, orientation, and WWR―were selected in compliance with the national energy efficiency standard for cold climates (JGJ26-2018), actual implementation may encounter several barriers. Local zoning regulations and detailed urban planning codes often impose more restrictive controls on building height, spacing, and orientation. For instance, some cities cap the height of residential towers or mandate fixed setback distances, which can hinder optimal solar access. From an economic perspective, mixed-height configurations and larger south-facing glazing areas, while beneficial to performance, may increase construction and façade costs, particularly in prefabricated or cost-sensitive housing projects. Moreover, occupant behaviors, HVAC operational constraints, and system maintenance issues―which were not modeled in the simulations―may diverge from the idealized settings, potentially diminishing the realized benefits of optimization. Future research should therefore investigate adaptive design strategies that balance regulatory, economic, and behavioral constraints to ensure both performance gains and practical viability.
6 Conclusions and future work
6.1 Summary
This research examines the rapid development of urban residential areas and the growing issue of energy consumption. Despite the significant impact of residential morphology on building energy consumption and performance, existing studies are often fragmented and lack systematic analysis. This study addresses that gap. By integrating theories of energy efficiency and residential design, we propose and validate a multi-objective optimization workflow based on the Grasshopper in the context of Qingdao’s cold-climate climate. This study investigates the effects of six typical residential morphologies in cold-climate regions on building energy consumption, daylight, and thermal comfort via MOO and quantitive analysis. Utilizing the Pareto optimal solution method, we optimized design variables such as OB, HB, and WWR. The findings reveal that reasonable adjustments to these variables can significantly enhance building performance, specifically optimizing EUI, UDI, and PPD values.
The research results indicate that the Pareto optimal solutions achieved a maximum reduction of 6.14% in EUI, a maximum increase of 7.92% in UDI, and a maximum reduction of 4.84% in PPD. These improvements were obtained through the optimization of design variables such as the WWR-S, HB, and spatial form. For example, increasing WWR-S reduces energy consumption and improves thermal comfort, whereas increasing WWR-WE improves thermal comfort but may decrease daylight performance. Among the layout types, the Slab Layout demonstrated superior energy performance compared to both the Point Layout and Point-Slab Hybrid Layout, while Mixed Height vertical distributions performed better in daylighting than Highrise Only schemes. Slab-Highrise emerged as the most robust form across all metrics, offering statistically and practically balanced performance.
The six model types utilized in this study were created from actual survey data and adhere to local planning standards, guaranteeing engineering relevance in contrast to many hypothetical optimization studies. The chosen variables enhance the findings’ practical usefulness by representing typical early-stage design decisions. Additionally, the model’s reliability was further confirmed by validating simulation outputs using empirical energy consumption data from a government-led investigation.
Although the absolute numerical improvements in EUI, UDI, and PPD remain moderate, they represent reliable optimization outcomes under early-stage assumptions. Given the scale and simplification of district-level modeling, the consistency of improvement trends across all models, as well as their validation through statistical tests, reinforces the directional robustness and practical relevance of the findings. This gives practitioners confidence when navigating intricate trade-offs in early planning by highlighting the importance of morphological optimization even in the presence of modeling tolerance.
Combining Pareto trend analysis with non-parametric statistical comparisons, this dual-track evaluation allows for both generalization across broader design contexts and accuracy in scenario-based decision-making. By bridging form-based strategies with validated performance modeling, this study contributes a transferable methodology for guiding energy-efficient residential planning in cold-climate regions, with potential applicability to other climatic zones.
6.2 Practical and theoretical significance
This study establishes a scientifically grounded and practically applicable framework for optimizing residential district morphology in cold-climate regions. The contributions are twofold:
From a practical standpoint, the selected spatial forms are not hypothetical but derived from surveys of 660 real-world residential districts in Qingdao. The models strictly adhere to local planning regulations and national design standards (e.g., GB 50096), ensuring that results are aligned with real engineering constraints. Furthermore, the design variables―plan layout, building orientation, height (HB), and window-to-wall ratio (WWR)―are among the most influential and adjustable parameters during early planning stages. This makes the findings directly valuable for designers and decision-makers seeking to improve energy performance and environmental quality during the schematic design phase. The model’s assumptions were further validated and its reliability was increased by comparing the simulation findings to actual energy-use data from a government-sponsored study.
The study theoretically fills the gap between the evaluation of energy performance and morphological typology. It combines non-parametric statistical analysis, parametric modeling, and MOO into one framework. The results are accurate for high-performing scenarios and generalizable to larger metropolitan contexts due to the application of both population-based statistical testing and Pareto-based solution evaluation. In addition to supporting future research examining integrated passive-active design systems, this dual-track framework offers a methodological basis for developing performance-driven design methods.
In summary, this study contributes to both the methodological development and practical application of early-stage residential optimization, providing scientific evidence and transferable strategies to guide sustainable urban planning in cold climates―and potentially in other climate zones with similar design challenges.
6.3 Future work
Future research will expand the optimization framework in three key directions: performance dimensions, practical constraints, and system integration.
First, the scope of optimization objectives will be broadened to include additional factors such as glare control, ventilation, indoor air quality, acoustic comfort, and cost considerations. Expanding the range of objectives will provide a more comprehensive evaluation of building performance. In particular, incorporating additional sustainability-related metrics such as embodied carbon, lifecycle environmental impacts, and detailed cost analysis would further strengthen the robustness and comprehensiveness of the optimization framework. Although these dimensions require more extensive data typically available at later design stages, future research should explore integrating lifecycle assessment (LCA) methodologies and detailed cost metrics when project-specific information becomes accessible. As the number of objectives increases, advanced algorithms such as NSGA-III and Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) will be adopted to efficiently manage higher-dimensional trade-offs.
Second, future work will specifically include project-specific limitations, such as developer requirements, market-driven typologies, and adherence to local and national building codes, to improve real-world applicability. To preserve tractability throughout the early stages of design, these factors have been simplified in the current work. In addition, optimization tools will be further developed to integrate with widely used planning platforms such as AutoCAD and GIS, improving usability for designers and policymakers.
Third, beyond morphological parameters such as layout, orientation, and window configuration, future research will incorporate envelope-level variables, such as insulation thickness, glazing types, similar to NSGA-II optimizations of envelope properties and dynamic shading in office buildings (
liu et al., 2024), and thermal mass, to allow for more holistic passive design strategies. Moreover, active system-level measures such as adaptive HVAC controls, occupancy-based thermal zoning, and daylight-responsive lighting will be introduced. These integrations aim to enhance the realism of energy and comfort predictions, bridging the gap between early-stage form generation and operational performance assessment.
2095-2635/2025 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd.
PDF
(5182KB)
