Hybrid metaheuristic and DL approach for integrated solar and electric vehicle energy management in smart grid networks

Abstract

The increasing penetration of electric vehicles (EVs) and intermittent renewable generation has created challenges for accurate energy forecasting and stable smart-grid operation. Existing forecasting approaches often retain redundant variables or use manually selected model parameters, limiting their predictive reliability. This study developed a deep representation learning-based integrated solar and EV energy management (DRL-ISEVEM) framework to improve EV energy-consumption forecasting for smart-grid decision support. The input data were standardized using Z-score normalization. An improved whale optimization algorithm (IWOA) subsequently selected 17 informative variables from 19 candidate features, while a multivariate temporal convolutional network (M-TCN) captured nonlinear temporal dependencies and inter-variable relationships. A dung beetle optimizer (DBO) tuned the M-TCN hyperparameters by minimizing the validation error. Experimental evaluation on a benchmark EV energy-consumption dataset produced testing values of 0.001308 for mean squared error (MSE), 0.036168 for root mean squared error (RMSE), 0.029003 for mean absolute error (MAE), and 0.05438 for mean absolute percentage error (MAPE). The testing RMSE was only 1.46% higher than the training RMSE of 0.035648, indicating a small within-dataset generalization gap. These findings demonstrated that the sequential integration of feature selection (FS), temporal learning, and hyperparameter optimization provided accurate EV energy forecasts for smart-grid energy-planning applications.