Optimizing household energy management with distributed energy resources: A multi-learning-guided stochastic optimization approach

Modern households, installed with distributed energy resources such as renewable energy systems and storage units, can self-consume the generated energy, sell surplus energy to the grid, or combine both approaches based on the operational context. Optimizing these energy resources is essential for reducing costs and improving the efficiency of energy usage in households. This paper introduces an optimization of household energy resources model managed by an energy service provider. The study targets residences with technologies aimed at reducing energy costs through effective Demand Response (DR) activities. A precise model is formulated to determine the optimal schedule for home appliances, focusing on minimizing energy costs and reducing DR curtailment activities. In addition to the scheduling model, the paper features a new approach using the Fast Random Opposition-based Learning with Reinforced Learning Grey Wolf Optimizer (F2R-GWO) to solve the optimization problem through two methods: an integrated approach, where the variables of all households are combined into a single optimization framework, and a distributed method, which takes advantage of the independence among household parameters by utilizing a multi-population strategy. A multi-population strategy involves dividing the population into subgroups to enhance the diversity of the algorithm. The study also utilized real-time household energy consumption datasets including electricity pricing and renewable energy integration. The results indicate that the distributed approach significantly enhances the efficiency of the F2R-GWO, particularly when dealing with larger datasets. This is especially evident when the problem dimension increases with the number of households, making the F2R-GWO more beneficial. This study also evaluates the proposed algorithm against five methods, including Linear-Success-History Adaptation Differential Evolution with Semi-Parameter Adaptation hybrid with Covariance Matrix Adaptation evolution strategy (LSHADE-SPACMA), neighbourhood mutation operators and opposition-based learning differential evolutionary algorithm, butterfly optimization algorithm with phasmatodea population evolution, Multimodal Delayed Particle Swarm Optimization (MDPSO) algorithm, and GWO. The proposed F2R-GWO algorithm achieved a 3.5 % reduction in daily energy costs compared to LSHADE-SPACMA and demonstrated a 9.4 % improvement in convergence efficiency over MDPSO. The proposed algorithm consistently outperforms other algorithms, achieving better fitness values in all test case studies, thus demonstrating its superior capability.