A modern survey method for determining live loads based on multi-source and open-access data on the Internet

Abstract

Sufficient survey data are required to describe the stochastic behaviors of live loads. However, due to manual and on-site operation required by traditional survey methods, traditional surveys face challenges like occupant resistance, high costs, and long implementation periods. This study proposes a new survey method to access live load data online and automatically. Required samples are acquired from multi-source, open-access and dynamically updated data on the Internet. The change intervals, geometrical dimensions and object quantities are obtained from transaction information, building attributes and virtual reality models on real estate websites, respectively. The object weights are collected from commodity information on e-commerce websites. The integration of the aforementioned data allows for the extraction of necessary statistics to describe a live load process. The proposed method is applied to a live load survey in China, covering 20040 m², with around 90000 samples acquired for object weights and load changes. The survey results reveal that about 70%−80% of the amplitude statistics are attributable to 1/6 of the total object types.

Graphical abstract

Keywords

live load / online survey method / data acquisition / data integration

1 Introduction

Engineering structures are designed to meet target reliability levels. In structural reliability analysis, the loads are generally the most uncertain factors [1]. Sufficient load data are required to estimate the parameters of available models and facilitate the development of innovative models. This study focuses on live load data acquisition.

The live load is one of the most basic structural loads due to the occupancy of structures [2]. It is classified into the sustained load and extraordinary load according to temporal behaviors [3]. The sustained load produced by normal furniture and personnel is approximated as constant until an occupant change [4]. The extraordinary load is caused by unusual events such as crowd gathering and furniture stacking [5], and is characterized by low occurrence frequencies and short duration. The survey method for the sustained load is of interest in this study. The investigation into the extraordinary load is considered elsewhere [6].

According to the probabilistic model describing the live load [7,8], statistical properties of load amplitudes and change intervals are required in the investigations of the sustained load. Manual and on-site investigations employed by traditional survey methods lead to resistance from occupants, low efficiency and high costs.

When investigating load amplitudes, investigators require to physically enter the surveyed regions and weigh all the indoor objects. As a simplification, the inventory technique [9] indirectly estimates the weight of an object according to its type, material and dimensions. This technique is adopted by several researchers [10–12] but is considered an approximate method due to the difficulties with inferring weight from limited information. Moreover, such load investigations still require to be conducted on site.

When investigating change intervals, investigators need to interview occupants on site [13,14], trace addresses through telephone directories [12] or search statistical reports by governments [15] to obtain related data. Conducting interviews can lead to occupant resistance, long implementation periods and unreliable results. The telephone directory information is difficult to obtain when the occupants have high privacy requirements. Inaccessibility to detailed and raw data from official reports impedes further analyses of load behaviors.

Taking full advantage of unlimited data on the Internet, this study proposes a new survey method for determining live loads. Virtual reality models are utilized to acquire object quantities, residence attributes provide geometric dimensions and occupant behaviors, and commodity information gives object weights. Subsequently, the required load statistics can be obtained through an integration of multi-source online data. Unlike traditional methods, data acquisition and processing are entirely conducted online, significantly enhancing the feasibility and efficiency of the investigation. In addition, leveraging open-access data for load statistics can substantially reduce the survey cost.

The remainder of this paper is organized as follows. Section 2 introduces the statistical properties necessary for a load description and the corresponding data types to be obtained in load investigations. Section 3 illustrates the acquisition and integration process of Internet data used for determining load amplitudes and change intervals. Section 4 applies the proposed method to a live load investigation of residential buildings in Shanghai, China.

2 Data requirements of load investigations

A Poisson square process is generally employed to describe temporal behaviors of sustained loads [7,8]. Then, the sustained load at time t can be expressed as:

(1)S(t)=\sum _{i=0}^{N(t)}{\theta }_i{I}_i\left(t,{\tau }_i,{\tau }_{i+1}\right),

where N(t) is the total number of occupant changes within t; {\tau }_i stands for the time of the ith occupant change and equals zero for i = 0; {\theta }_i represents the load amplitude between {\tau }_i and {\tau }_{i+1}; I_i\left(t,{\tau }_i,{\tau }_{i+1}\right) is given by:

(2)I_i\left(t,{\tau }_i,{\tau }_{i+1}\right)=\{\begin{array}{ll}1,& t\in \left[{\tau }_i,{\tau }_{i+1}\right],\\ 0,& t\notin \left[{\tau }_i,{\tau }_{i+1}\right].\end{array}

Fig.1 presents a typical sustained load process. The change interval in Fig.1 is defined as the length of time between two adjacent occupant changes and is denoted by T_{\mathrm{c}} in this study. According to the Poisson process model, T_{\mathrm{c}} is a random variable following the exponential distribution. The mean T_{\mathrm{c}} is given by:

(3)E\left(T_{\mathrm{c}}\right)=\frac{1}{\lambda },

where \lambda denotes the mean change rate of the sustained load process.

The exponential distribution only contains one distribution parameter and can be completely determined from \lambda. As shown in Fig.1, the mean change interval is employed to estimate \lambda in load investigations.

Moreover, the load amplitude in Fig.1 is required to describe a sustained load process. The load amplitude can be represented by a random variable based on the theoretical analyses of spatial variability [8]. The unit load and equivalent uniformly distributed load are generally employed to represent the load amplitude [16–18]. The equivalent uniformly distributed load is defined to produce the same load effect as actual loads, which depends on mechanical analyses of specific structures. This study focuses on the acquisition of load statistics and adopts the unit load to represent the load amplitude.

The unit load is defined as:

(4)U=\frac{{\iint }_{A}v(x,y)\mathrm{d}x\mathrm{d}y}{A},

where v(x,y) represents the load intensity at point (x,y) and A is the area of the region of interest.

v(x,y) is assumed to obey the following form [19]:

(5)v(x,y)=Y+\epsilon (x,y),

where Y is a random variable describing the average unit load on a floor and \epsilon (x,y) represents a random process with zero mean describing the spatial fluctuation relative to Y.

The mean U is given by [19]:

(6)E(U)=m,

where m is the mean of all the unit loads for a particular occupancy type (e.g., office buildings).

The load intensities of two points separated by any finite distance are assumed to be uncorrelated [8]. The area-dependent variance of U is expressed as the following:

(7)D(U)={\sigma }_1^2+\frac{{\sigma }_2^2}{A},

where {\sigma }_1^2 stands for the variance in individual floor averages from m; {\sigma }_2^2 is an experimental constant [19].

The mean and variance of a load amplitude are sufficient to determine its distribution parameters, because a distribution containing two parameters is typically adopted. Equations (6) and (7) suggest that the mean and variance of a load amplitude will be available if m,{\sigma }_1 and {\sigma }_2 are obtained. As illustrated in Fig.1, the mean of U is used for estimating m. The relationship between variances of U and areas of surveyed regions is employed to estimate {\sigma }_1 and {\sigma }_2.

The probabilistic description of a sustained load process is completely determined if the distributions of its load amplitude and change interval are available. Subsequently, the interested characteristics of the sustained load process can be calculated. For example, the maximum load in the reference period presented in Fig.1 is of interest in the determination of design live loads [4] and has the probability density function (PDF) of [15]:

(8)f_{L_{\mathrm{m}\mathrm{a}\mathrm{x}}}(l)=\exp \left\{-\lambda T\left[1-F_{\mathrm{L}}(l)\right]\right\}\left[1+\lambda T{F}_{\mathrm{L}}(l)\right]f_{\mathrm{L}}(l),

where f_{\mathrm{L}}(l) represents the PDF of the load amplitude; F_{\mathrm{L}}(l) is the cumulative distribution function (CDF) of the load amplitude; T stands for the reference period.

3 Online survey method

The traditional method requires load investigators to physically enter the surveyed regions and carry out a series of operations, such as measuring room areas, determining object weights and asking the occupants about their moving experience. These on-site operations will consume significant manpower and time. Moreover, occupants are very likely to worry that an on-site investigation will invade their privacy and disrupt their normal living or working routine.

To address these problems, an online survey method is developed in this study.

3.1 Investigation framework

The schematic diagram of the proposed method is illustrated in Fig.2. Multi-source heterogeneous data on the Internet are utilized to obtain required samples.

Massive data with high-dimensional characteristics are available on the Internet. According to data requirements of load investigations, three types of data are extracted from real estate websites. Transaction information provides the date of occupant changes in the surveyed regions, building attributes contain geometrical dimensions used for calculating room areas, and virtual reality models allow the counting of indoor objects without entering surveyed regions. Moreover, commodity information including object weights is extracted from e-commerce websites.

The statistics for load amplitudes can be obtained through integrating room areas, object quantities and object weights, as detailed in Subsection 3.2. The samples and corresponding statistics for change intervals are acquired by analyzing occupant behaviors, as detailed in Subsection 3.3.

Subsequently, the properties of interest of a sustained load process can be derived based on the probabilistic model in Section 2.

3.2 Investigation into load amplitude

As mentioned above, the mean and variance of the unit load are required to determine the distribution of the load amplitude. The unit load in the surveyed region can be expressed as:

(9)U=\frac{{\displaystyle \sum _{j=1}^R{W}_{j,1}+W_{j,2}+\cdots +W_{j,{\eta }_j(A)}}}{A},

where A stands for the area of the surveyed region; R is the total number of indoor object types; {\eta }_j(A) represents the quantity of the jth type object over A; W_{j,k}[k=1,2,\dots ,{\eta }_j(A)] is the piece weight of the kth object belonging to the jth type.

The piece weights of different objects of the same type are assumed to be independent and follow the same distribution. The total weights of different types of objects are assumed to be independent of each other.

The mean of the unit load is given by:

(10)E(U)=\sum _{j=1}^R\frac{E\left[{\eta }_j(A)\right]E\left(W_j\right)}{A},

where W_j is the piece weight of an object belonging to the jth type; E\left(W_j\right) is the mean of W_j; and E\left[{\eta }_j(A)\right] represents the mean of {\eta }_j(A).

The variance of the unit load reads:

(11)D(U)=\sum _{j=1}^R\frac{E\left[{\eta }_j(A)\right]D\left(W_j\right)+D\left[{\eta }_j(A)\right]E^2\left(W_j\right)}{A^2},

where D\left(W_j\right) and D\left[{\eta }_j(A)\right] are the variances of W_j and {\eta }_j(A), respectively.

Equations (10) and (11) suggest that the total number of indoor object types to be considered requires to be determined. Moreover, the statistics describing the quantity and piece weight of each object type are necessary to be obtained. This study acquires all the above information from the open access data on real estate and e-commerce websites. The investigation process is illustrated in Fig.3.

Moreover, the contribution of each object type to load amplitude statistics can be quantified based on the proposed method. Equations (10) and (11) reveal that the mean and variance of the load amplitude are both the sum of R items. Each item represents the contribution of an object type.

3.2.1 Determination of the total number of object types

Some common types of indoor objects can be listed first. Subsequently, the list of object types is gradually improved in the investigation. The object types in each surveyed room are identified from its virtual reality model, as described in Subsubsection 3.2.2. If the type of an object is not included in the original list, it is added to the list. The parameter R in Eqs. (10) and (11) is taken to be the total number of object types in the list after all the rooms are investigated.

3.2.2 Acquisition of quantity statistics and corresponding areas

This section introduces the acquisition of the quantity {\eta }_j(A) and its corresponding area A, as in Eqs. (10) and (11). Specifically, the area and object quantities of a surveyed room require to be obtained.

Real estate websites provide multi-source heterogeneous data on properties for sale. The structured and unstructured data are available online for potential buyers to reference, as shown in Fig.4. The structured data includes housing prices, room areas, floor locations and other related information. Therefore, the area of a surveyed room can be directly obtained.

The unstructured data contains room photos, room videos and virtual reality models. The photos and videos may miss some indoor objects, but a virtual reality model allows load investigators to explore every corner of a surveyed room without being on site. The observations of indoor objects based on the virtual reality model of a real estate website (available at the website of Lianjia) are illustrated in Fig.5. Load investigators can access the virtual reality model of a surveyed room and record the type and quantity of the observed indoor objects. Moreover, a room is investigated online by different investigators and the results are cross-validated to ensure data quality. If the object quantities obtained by different investigators are different for the same room, this room needs to be re-surveyed until consistent results are acquired.

3.2.3 Acquisition of piece weight statistics

This section focuses on the acquisition of the weight W_j, as in Eqs. (10) and (11). In other words, the samples of the piece weights need to be obtained for each object type.

With the growing popularity of online shopping, e-commerce websites have gathered a huge amount of commodity data. Fig.6 presents some of the commodity information on an e-commerce website. The structured data generally includes prices, dimensions, piece weights and other information of the commodity. The unstructured data consists of the brands, materials, commodity photos and so on. Therefore, the piece weights concerned in this study can be directly obtained from the structured data. Moreover, the boxplot method [20] is employed in the data cleaning of weight samples.

3.3 Investigation into change intervals

The mean change interval requires to be obtained to estimating the mean change rate of the sustained load process. This study acquires the mean change interval based on the transaction information of properties on real estate websites. The data structure describing the transaction information on a real estate website is presented in Fig.7. Different types of data are provided for the properties on sale and the properties sold separately.

The dates of last transaction and listing for sale are available for the properties on sale. The period between the date of last transaction and the date of listing for sale is defined as the occupied period. In other words, the occupied period is the time between when an occupant moves into a property and when he or she lists this property for sale.

The sales period in Fig.7 is defined as the period between when an occupant lists his or her property for sale and when he or she sells it to the next occupant. Therefore, the sales period is only available for the properties that have been sold.

Fig.8 presents the process of occupant changes in a property and illustrates the definition of the occupied and sales period. The time of transaction can be approximated as the time of the occupant change. Therefore, the change interval T can be expressed as:

(12)T=T_{\mathrm{o}}+T_{\mathrm{s}},

where T_{\mathrm{o}} and T_{\mathrm{s}} represent the occupied period and sales period, respectively.

The mean change interval yields:

(13){\mu }_T={\mu }_{T_{\mathrm{o}}}+{\mu }_{T_{\mathrm{s}}},

where {\mu }_{T_{\mathrm{o}}} and {\mu }_{T_{\mathrm{s}}} are the means of the occupied period and sales period, respectively.

The samples of the occupied period and sales period can be acquired from the data of the properties on sale and the properties sold, respectively. The boxplot method [20] is employed in the data cleaning process of the occupied period and sales period. Subsequently, the mean change interval can be derived based on the means of the occupied period and sales period.

The occupied period covers most of the time that an occupant lives or works in a property. The sales period refers to the time taken by the occupant to find a buyer. The mean sales period is generally much smaller than the mean occupied period. Therefore, the influence of the sales period can be neglected in load investigations, which is verified by subsequent survey results.

3.4 Comparison with traditional methods

The acquisition of live load data based on the traditional and proposed method is compared in Tab.1.

Load investigators do not need to enter the surveyed regions and most data can be automatically and online obtained based on the proposed method. Due to the dynamic update of Internet data, the raw data of the proposed method can be generated continuously without increasing survey costs.

4 Load investigations based on proposed survey method

4.1 Basic information

The proposed method is applied to a live load survey of residential buildings in Shanghai, China. As mentioned in Subsection 3.1, the acquisition of load amplitudes and change intervals are two basic survey tasks. Subsection 4.2 obtains load amplitude statistics from an investigation into 20040 m² of floor areas and 44878 weight samples. Subsection 4.3 accesses the change interval statistics based on 12636 samples of the occupied period and 30004 samples of the sales period. Subsequently, Subsection 4.4 yields the probability distribution of the maximum sustained load in the reference period grounded in the above two statistics.

4.2 Investigations into load amplitudes

The object quantities of 1007 living rooms covering 20040 m² in Shanghai are acquired. The quantities are identified from the virtual reality models of these rooms on a real estate website (available at the website of Lianjia). The areas of these rooms are also accessible on this website. According to theoretical analyses in Section 2, the variance of the load amplitude is area dependent. Therefore, the surveyed rooms are classified into different groups according to their areas. The area ranges and numbers of rooms for each group are provided in Tab.2.

A total of 45 types of indoor objects are considered. The mean and standard deviation (SD) for the quantities of some typical types are presented in Fig.9. The quantity statistics of different object types present various tendencies with increasing area. The quantity statistics of the sofa and chair basically increase with increasing area. Larger living rooms are often used to host more guests or larger family gatherings and therefore usually require more seating. However, the quantity statistics of the fridge generally decrease with increasing area. Placing a fridge in the kitchen allows for easier food storage and cooking. In situations where space is limited, such as in small apartments, the fridge has to be placed in the living room. Living rooms in smaller homes are also generally smaller in size and therefore more likely to contain a fridge.

A total of 44878 samples of piece weights are obtained from an e-commerce website (available at the website of Jingdong). The mean, SD and sample size of some typical types are listed in Tab.3. Moreover, the residents also contribute to the sustained load of living rooms. The number of residents in a living room is taken as the number of people in a family. The mean and SD describing the number of people are taken to be 2.62 and 0.79, respectively [21,22]. The weight of a person has the mean and SD of 62.5 and 10.0 kg, respectively [23].

Subsequently, the mean and variance of the unit load can be obtained from the weight statistics and quantity statistics. The mean unit load is 0.291 kN/m² and the SD of the unit load is presented in Tab.4. According to the least squares fitting, {\sigma }_1 and {\sigma }_2 in Eq. (7) are taken as 0.009 and 0.345, respectively. Therefore, the unit load has the variance of:

(14)D(U)=8.623\times {10}^{-5}+\frac{0.119}{A}.

The load amplitude is assumed to follow the extreme value type I distribution, whose distribution parameters can be determined based on the mean and variance. The probability distributions of the load amplitude for typical areas are presented in Fig.10. The evolution of the amplitude distribution with area is due to the change of the variance. The variation of typical fractile values describing the load amplitude with area is illustrated in Fig.11. All the fractile values decrease with increasing area due to the constancy of the mean and the decrease of the variance.

As mentioned in Subsection 3.2, the contribution of each object type to the amplitude statistics can be obtained according to Eqs. (10) and (11). The contribution of some predominant types to the mean amplitude is presented in Fig.12. The personnel are also considered as a type of indoor objects. The contribution to the variance of the amplitude is illustrated in Fig.13.

Fig.12 and Fig.13 suggest that the contributions of different objects are extremely unbalanced. 7 types of indoor objects account for around 75% of the mean and 70% of the variance. The remaining 38 types make up a relatively low proportion. The personnel and sofa are the top two types in terms of their contributions to the amplitude statistics. The contribution of a particular type to the amplitude variance can be different for different groups, which is due to the change of quantity statistics with area.

4.3 Investigations into change intervals

The statistical properties of occupant changes in residential buildings are significantly influenced by local housing prices. Therefore, the samples related to change intervals are obtained for different districts separately. Nine districts in Shanghai are considered and their sample sizes of the occupied period and sales period are displayed in Tab.5 and Tab.6, respectively. The mean occupied period and sales period are illustrated in Fig.14.

Fig.14 suggests that the mean occupied period is one to two orders of magnitudes larger than the mean sales period. Consequently, the mean occupied period can be used to approximate the mean change interval.

As mentioned in Subsection 3.3, the transaction interval is employed to approximate the change interval. High housing prices will increase the cost of transactions, resulting in a longer transaction interval. Therefore, the districts with higher housing prices should have longer means of change intervals. As illustrated in Tab.7, the mean change intervals of nine districts are classified into types I and II based on K-means clustering [24]. The mean change intervals of the districts in type I and type II are 10.18 and 8.35 years, respectively. The probability distributions of the change intervals are displayed in Fig.15. It is found that all the districts that belong to the central city of Shanghai and have higher housing prices [25] are classified as type I, which indicates the consistency between the survey results and the economic background.

4.4 Maximum load in the reference period

Based on Eq. (8), the probability distribution of the maximum sustained load in the reference period can be determined from the amplitude statistics in Subsection 4.2 and the interval statistics in Subsection 4.3. The samples of the change interval in nine districts are combined to obtain a unified mean interval of Shanghai. The reference period and mean change interval are taken as 50 years and 9.03 years, respectively. The probability distributions of the maximum load for some typical areas are presented in Fig.16. The variation of the mean, SD and typical fractile values describing the maximum load with area is displayed in Fig.17. The mean change rate and mean amplitude both remain constant with the change of area. The decrease of load values with increasing area in Fig.17 is caused by the decrease of the amplitude variance.

5 Conclusions

A modern survey method for determining live loads is proposed. The required samples of load investigations are acquired online from the open-access data on real estate and e-commerce websites, providing high efficiency, low cost, and dynamic update.

The proposed method is applied to a live load survey of residential buildings. 20040 m² of floor areas, 44878 samples of piece weights and 42640 samples related to change intervals are covered. The contribution of each object type to the load amplitude statistics is quantified based on survey data. The contributions of different types are significantly unbalanced. 7 out of 45 object types contribute about 75% of the mean and 70% of the variance. Survey results suggest that the mean amplitude and mean change interval are 0.29 kN/m² and 9.03 years, respectively.

Based on the currently proposed method, manual operation is necessary only for counting the quantity of objects, while automatic online acquisition has been successfully implemented for other types of data. In future research, the consideration of applying computer vision technology could enable the automatic recognition of object quantities, thereby achieving complete automation of the entire survey method.

References

[1] MelchersR EBeckA T. Structural Reliability Analysis and Prediction. Chichester: John Wiley & Sons, 2018

[2] EllingwoodBGalambosT VMacGregorJ GCornellC A. Development of A Probability Based Load Criterion for American National Standard A58 Building Code Requirements for Minimum Design Loads in Buildings and Other Structures. Washington, D.C.: National Bureau of Standards, 1980

[3] Corotis R B, Jaria V A. Stochastic nature of building live loads. Journal of the Structural Division, 1979, 105(3): 493–510

[4] Chalk P L, Corotis R B. Probability model for design live loads. Journal of the Structural Division, 1980, 106(10): 2017–2033

[5] Choi E C. Extraordinary live load in office buildings. Journal of Structural Engineering, 1991, 117(11): 3216–3227

[6] Xu C, Chen J, Li J. A modern survey method for determining extraordinary loads based on multi-source online data. Advances in Structural Engineering, 2023, 26(11): 2123–2134

[7] Peir J C, Cornell C A. Spatial and temporal variability of live loads. Journal of the Structural Division, 1973, 99(5): 903–922

[8] McGuire R K, Cornell C A. Live load effects in office buildings. Journal of the Structural Division, 1974, 100(7): 1351–1366

[9] Culver C G. Live-load survey results for office buildings. Journal of the Structural Division, 1976, 102(12): 2269–2284

[10] Harris J C, Corotis R B. Hospital inventory load survey. Journal of the Structural Division, 1978, 104(12): 1859–1868

[11] Andam K A. Floor live loads for office buildings. Building and Environment, 1986, 21(3–4): 211–219

[12] Kumar S. Live loads in office buildings: Point-in-time load intensity. Building and Environment, 2002, 37: 79–89

[13] NiuJ GNiuD T. Investigation and statistical analysis on the floor loads of residential buildings. Journal of Xi’an University of Architecture & Technology (Natural Science Edition), 2006, 38(2): 214–220 (in Chinese)

[14] GeS JChenHSunZ SLiJ B. Survey and statistics of floor live load of residential building in central plains region. Building Structure, 2008, 38(7): 88,125–128 (in Chinese)

[15] Harris M E, Bova C J, Corotis R B. Area-dependent processes for structural live loads. Journal of the Structural Division, 1981, 107(5): 857–872

[16] Choi E C C. Live load model for office buildings. The Structural Engineer, 1989, 67(24): 421–425,437

[17] Ruiz S E, Soriano A. Design live loads for office buildings in Mexico and the United States. Journal of Structural Engineering, 1997, 123(6): 816–822

[18] Kumar S. Live loads in office buildings: Lifetime maximum load. Building and Environment, 2002, 37: 91–99

[19] Ellingwood B, Culver C. Analysis of live loads in office buildings. Journal of the Structural Division, 1977, 103(8): 1551–1560

[20] LiHHuQ Y. Forecasting and Decision Making. Beijing: China Machine Press, 2019 (in Chinese)

[21] NationalBureau of Statistics of the People’s Republic of China. The Seventh National Census Bulletin (No. 2). 2021 (available at the website of National Bureau of Statistics of the People’s Republic of China) (in Chinese)

[22] HongT H. Harmonious Family Building in China. Beijing: Social Sciences Academic Press, 2011 (in Chinese)

[23] Standardizationadministration of the People’s Republic of China. Human Dimensions of Chinese Adults (Draft for Comment). 2020. Available at the website of National public service platform for standards information (in Chinese)

[24] TanP NSteinbachMKumarV. Introduction to Data Mining. Beijing: China Machine Press, 2010

[25] ShanghaiBureau of Statistics. Shanghai’s Seventh National Population Census Main Data Bulletin (No. 2). 2021. Available at the website of Shanghai Bureau of Statistics (in Chinese)

RIGHTS & PERMISSIONS

Higher Education Press