Hybrid Reinforcement Learning for Adaptive PSO-ANFIS in University Energy Consumption Forecasting

Adequate energy management and planning require accurate consumption forecasting. This study developed a hybrid reinforcement learning model to adaptively change the parameters of PSO based on a time-varying acceleration coefficient (RPSO) combined with adaptive network-based fuzzy inference systems (ANFIS) called the (RPSO-ANFIS) model, where reinforcement learning adaptively adjusts PSO parameters using a time-varying acceleration coefficient (RPSO). The model was compared with standalone ANFIS and PSO-ANFIS for forecasting electricity consumption in university residences from September 1 to December 31. The obtained result demonstrates the performance of the RPSO-ANFIS, which yields the minimum values in all performance metrics, with values of 1.6232 for RMSE, 1.2846 for MAD, 39.509% for MAPE, and 1.2922 for MAE. The results were compared with the ANFIS model, which gives 1.7224, 1.3792, 47.4645%, and 1.3745 for RMSE, MAD, MAPE, and MAE, respectively. Also, the hybrid PSO-ANFIS model offers 1.6929, 1.357, 43.6565%, and 1.3645 for RMSE, MAD, MAPE, and MAE, respectively. These findings confirm that RPSO-ANFIS effectively tunes ANFIS parameters, providing more accurate energy forecasts and supporting improved future energy planning and management.