Hybridizing Metaheuristic Methods with AI for Accurate Prediction of Residential Energy Consumption

Accurately predicting Total Energy Consumption in the Residential Sector is essential for sustainable energy planning. This study introduces four hybrid metaheuristic-artificial intelligence models—Earthworm Optimization Algorithm (EWAMLP), Stochastic Fractal Search (SFSMLP), Vortex Search (VSMLP), and Shuffled Complex Evolution (SCEMLP)—to enhance prediction accuracy. The models were evaluated using Root Mean Squared Error (RMSE) and R-squared (R2) on training and testing datasets, with swarm sizes optimized for each method. SFSMLP achieved the best overall performance, ranking first with the highest R2 values of 0.9936 (training) and 0.9859 (testing) and the lowest total score. VSMLP, with R2 values of 0.9915 (training) and 0.9866 (testing), tied for first in accuracy while demonstrating efficiency with an optimal swarm size of 300. SCEMLP, despite using the smallest swarm size (50), maintained competitive accuracy with R2 values of 0.9911 (training) and 0.9842 (testing), ranking third overall. EWAMLP, with R2 values of 0.9673 (training) and 0.9570 (testing), showed reliable but slightly lower performance, ranking fourth. These findings highlight the potential of hybrid metaheuristic-AI models for precise energy consumption predictions. The superior performance of SFSMLP and VSMLP suggests their suitability for applications requiring high accuracy, while SCEMLP offers a balance of efficiency and reliability. This study provides a robust framework for energy modeling, contributing to advancements in residential energy management and sustainability.