This paper presents an artificial intelligence (AI)-driven framework to enhance the reliability and efficiency of electric vehicle charging infrastructure through integrated anomaly detection and hybrid load forecasting. The methodology applies Autoencoders, isolation forest (IF), one-class support vector machine (OC-SVM), and density-based spatial clustering of applications with noise (DBSCAN) to identify irregular charging patterns, sensor faults, and operational anomalies. Using real-world data from the charging station in ASM Terni, Italy, the models detected anomalies ranging from 0.67% to 2.36% of sessions. For short-term energy demand forecasting, the proposed hybrid approach, combining seasonal auto regressive integrated moving average (SARIMA) with random forest (RF) residual correction, demonstrated that anomaly filtration is critical for predictive stability. On the filtered dataset, the hybrid model achieved a mean absolute error (MAE) of 11.95 kWh, outperforming both the standalone SARIMA (13.08 kWh) and RF (12.12 kWh) models. These findings underscore the importance of integrating anomaly detection into forecasting models to improve demand predictions, optimize charging schedules, and enhance energy management in smart EV charging networks. The proposed approach provides valuable insights for charge point operators (CPOs) and distribution system operators (DSOs) in improving the operational resilience of EV infrastructure. • Integrated AI framework links anomaly detection with hybrid EV load forecasting. • Autoencoder and DBSCAN display superior robustness in high-anomaly scenarios. • Hybrid SARIMA-RF model reduces prediction error to 11.95 kWh MAE. • Anomaly filtration is critical for preventing overfitting in load forecasts. • Validated on real-world Italian utility data for operational resilience.
Ghoreishi et al. (Wed,) studied this question.