Dark fermentation process from low-cost renewable substrates for simultaneous wastewater treatment and hydrogen production (H₂) is suitable due to economic viability and environmental sustainability. This work explores the application of an innovative control strategy in a scale fermentation bioreactor designed for energy recovery from organic wastes. This approach not only promotes low carbon emissions but also offers significant potential for industrial application. Machine learning (ML) and optimization methods are used to model the nonlinear process and then, a neural predictive control (NPC) strategy to drive the system to its optimal operating order under varying influent conditions is developed. Predictive control uses the Newton-Raphson as the optimization algorithm and a multi-layer feedforward neural network for the state prediction. This strategy has demonstrated to be a viable algorithm for real-time control applications. First, experimental data from continuous dark fermentation are modeled using support vector machine (SVM) algorithm for response prediction and then, optimization algorithms are employed to identify the key parameters that enhance H₂ production. These optimal operating parameters are then used to create reference trajectory signals within a NPC scheme to achieve the optimal hydrogen production rate. The control strategy led to an HPR mean of 12.35 ± 1.2 NL H₂/L-d under pseudo-steady state with hydrogen content in the gaseous phase of 63 % v/v, and a maximum COD recovery of 90% ± 2.8%. The results demonstrate that this innovative control method can significantly improve the performance and efficiency of biogas plants, showing viability for large-scale industrial implementation.

This study explores the potential and impact of electricity cogeneration using Organic Rankine Cycle (ORC) integrated with small-scale biomass boilers within district heating systems. An analysis is conducted on a 3 MW_th biomass-fired district heating plant in southern Sweden. Process monitoring data, collected over a one-year period from the plant, serves as the basis for simulation and analysis. The study examines operational changes and fuel usage at a local level, together with an extension to a regional scale considering both short-term and long-term energy system implications. The results show that integrating a 200 kW_e ORC unit with the existing boiler having a flue gas condenser is cost-optimal and could cogenerate approximately 1.1 GWh electricity annually, with a levelized electricity cost of €64.4 per MWh. This is equivalent to a system power-to-heat ratio of 7.5%. From a broader energy system perspective, this efficient integration could potentially reduce CO₂ emissions by 234-454 tons per year when the saved energy locally is used to replace fossil fuels in the energy system, depending on how biomass is utilized and what type of fossil fuels are replaced. Increasing installed capacity of ORC unit to maximize electricity co-generation could result in a carbon abatement cost ranging from €204 to €79 per ton CO₂. This cost fluctuates depending on the installed capacity, operation of the ORC units, and prevailing electricity prices. The study highlights the trade-off between financial gains and CO₂ emission reductions, underscoring the complex decision-making involved in energy system optimization.

With industrial informatization, abundant data provides solutions for the digital design of methane-based hydrogen production. Catalytic methane decomposition (CMD) is a promising strategy for COx-free hydrogen production, with high-value carbon products generated. However, affected by various factors, the proper process parameters are challenge to be ascertained by the time-consuming experimental method. In this study, five machine learning methods were utilized for the precise prediction of methane conversion using Ni-based catalysts. Combined with SHAP method and univariate analysis method, XGBoost model with the best accuracy (with R² = 0.894, RSME = 7.724) was selected for the exploration of the reaction impact of active phase loading, support loading, and reaction conditions in methane convention, hydrogen production, carbon yield, and carbon quality. The result shows that methane conversion rate is mainly influenced by space velocity, reaction temperature, nickel loading, and methane percentage. Copper doping significantly affects carbon yield and its quality, and there is a strong bond between Ni and Al₂O₃, contributing the most to the reaction. This work would provide a guidance for the efficient catalyst design and effective hydrogen production.

The depletion of hydrocarbon reserves and the impact of global warming have posed significant challenges to the continued use of fossil fuels. Consequently, renewable energy sources have garnered substantial attention, with some countries now deriving a significant portion of their total energy needs from these alternatives. Among renewable sources, wind energy has been recognized as one of the most accessible and clean. However, it is imperative to evaluate wind power plants both technically and economically. This involves calculating the levelized cost of energy in comparison to fossil-based energy sources and predicting the minimum and maximum energy output over the long term. Achieving this requires long-term forecasts of wind speeds at specific locations, which involve complex mathematical modeling and computations typically performed by supercomputers. In this study, a data-driven machine learning model has been employed to predict wind speeds in Calgary over a 25-year period with minimal CPU time. Throughout the power plant's operational life, the optimal model was also used to calculate the annual energy production. The hybrid CNN-LSTM model demonstrated superior accuracy based on model accuracy metrics. Consequently, the levelized cost of energy produced by the plant was calculated at $0.09 per kWh, which is competitive within the Canadian electricity market. The investment reached a breakeven point in approximately six years, which is deemed acceptable.

This study proposes a hybrid network of adaptive neuro-fuzzy inference system (ANFIS) with genetic algorithm (GA) to predict the higher heating value (HHV) of municipal solid waste (MSW). To enhance the robustness and accuracy of the model and optimize its ability to capture the complex non-linear relationship in the MSW dataset, eight membership functions (MF)-type of the grid partitioning (GP) clustering approach were tested. Moreover, understanding the relative importance and contribution of different waste properties to HHV prediction is critical for improving the model's predictive capability and optimizing the waste-to-energy (WTE) process. To this end, the feature importance analysis of MSW input variables was carried out using the decision tree regressor with the Gini importance (GI) metrics to identify the most influential variable. Key waste properties, including ultimate analysis data, ash and moisture content were used as input variables for the model. The result shows that the GP-clustered GA-ANFIS model based on triangular-shaped MF-type (tri-MF) has the most accurate HHV predictions with Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Mean Absolute Deviation (MAD) values of 0.7642, 13.677, 1.5913 and 0.9821 at the training and 0.6364, 16.216, 1.2437 and 0.7821 at the testing stage. Feature importance assessment revealed ash content as the most important predictor of HHV based on GI-value of 0.519668 (about 50% cumulative importance). Additionally, sulphur and nitrogen, along with ash content, dominated the HHV prediction and exhibited the highest predictive power on HHV with about 80% cumulative importance. The robust integrated approach of hybrid neuro-fuzzy model, with decision tree-based feature importance assessment, offers an effective approach for enhancing the prediction of HHV of MSW. The outcome of the study enhances WTE systems, facilitating more efficient and sustainable energy recovery from MSW.

The single hole injector, known for its simple design and ease of measurement, is widely utilized in optical spray experiments; however, multi-hole injectors are commonly applied in real engine applications. The structural differences between the two leads to variations in spray characteristics. This series of studies, based on the principles of similarity and normalization, proposes a theory for the transformation of spray characteristics between different hole numbers injectors. The 1st report investigates the spray characteristics of different hole numbers injectors under super high injection pressure conditions. Using the Diffuser BackgroundImaging (DBI) method, the experimental pressure range covers 100-300 MPa. The research indicate that the single-hole injector exhibits a shorter initial injection delay, while the multi-hole injector demonstrates a more stable injection flow rate and greater penetration. At higher pressures, the velocity increase, especially at 300 MPa. Higher ambient density has a suppressive effect on spray tip velocity and alters spray morphology. Moreover, it was observed that while the initial spray velocity of the single-hole injector is relatively higher, the penetration of the multi-hole injector significantly exceeds that of the single-hole injector in the later stages. For multi-hole injectors, interactions between adjacent sprays lead to a relatively narrower spray angle. The ratio of spray angle to cone angle for both injectors remain nearly unaffected by changes in density and injection pressure. In general, the Naber and Siebers model is better suited for predicting penetration in single-hole injectors under conditions of high density and ultra-high injection pressure (200-300 MPa). This study not only highlights the distinctive spray characteristics under super high pressure conditions but also offers valuable theoretical foundations and experimental insights for optimizing diesel engine design.