Electric Vehicle Charging Demand Forecasting Using Progressive Graph Convolutional Networks With Waterwheel Plant Optimization

Abstract

Electric vehicle (EV) charging demand forecasting is vital for maintaining smart grid (SG) stability, enhancing energy management (EM), and supporting large-scale EV integration. Nevertheless, current forecasting techniques often fail to maintain high prediction accuracy as well as computing efficiency while capturing complex spatial and temporal dependencies. To overcome these drawbacks, this study presents a hybrid Progressive Graph Convolutional Network integrated with the Waterwheel Plant Algorithm (PGCN-WWPA) for EV charging demand forecasting. The PGCN learns spatiotemporal relationships from charging data, while the WWPA optimizes network weights to improve convergence, reduce prediction errors, and enhance model generalization. The proposed framework was evaluated using an EV charging dataset and compared with existing approaches, including the Physics-Informed Attention Graph Network (PIAGN), Artificial Neural Network-Adaptive Neuro-Fuzzy Inference System (ANN-ANFIS), Long Short-Term Memory Neural Network (LSTMNN), Generative Adversarial Network (GAN), and Heterogeneous Spatial–Temporal Graph Convolutional Network (HSTGCN). The obtained results show that the proposed technique achieved classification performance of 0.97 accuracy, 0.97 precision, 0.99 recall, 0.98 specificity, and 0.98 F1-score, while yielding the lowest prediction errors, with an RMSE of 2.43, MAE of 1.98, and MSE of 0.54. The computational time was reduced to 5.2 s, representing a 30.7% improvement over the strongest baseline, and the model achieved an 18.5% improvement in prediction accuracy compared with HATGCNN. Overall, the outcomes prove that PGCN-WWPA provides more accurate, faster, and scalable EV charging demand forecasting, offering an effective solution for intelligent EM and future SG applications.