Enhanced machine learning algorithm for predicting energy consumption in smart buildings

Energy consumption in buildings represents a substantial share of global energy use and underscores the need for intelligent solutions to improve efficiency. Accurate prediction of heating and cooling loads is essential for optimizing energy usage in smart buildings and supporting broader sustainability goals. SRSm is a new variant of the Special Relativity Search (SRS) algorithm that incorporates a mutation phase to increase population diversity, enhance exploration, and reduce the risk of local optima. This variant is used to optimize a neural network for the accurate prediction of heating and cooling loads. The proposed model is evaluated using several statistical performance metrics and compared with conventional and advanced optimization techniques. The results show that SRSm consistently achieves competitive predictive accuracy. The integration of the mutation mechanism improves predictive performance and leads to high accuracy in both heating and cooling load estimation during the testing phase.