ML-Guided Coordinated EV Charging: LSTM Forecasting and Multi-Objective Optimization for Real-Time Grid Operation

Abstract

In this paper, an integrated real-time framework for coordinated Electric Vehicle (EV) charging is proposed based on load forecasting and multi-objective optimization techniques. The framework integrates load forecasting using Long Short-Term Memory (LSTM) and multi-objective optimization to minimize peak load, charging cost, and grid stress while maintaining high user satisfaction under dynamic smart-grid conditions. The proposed framework combines multi-step LSTM forecasting with a convex optimization scheduler operating on a 15-min rolling horizon that incorporates feeder constraints, electricity tariffs, charger limits, and departure state-of-charge requirements. Unlike conventional charging strategies, the proposed method enables adaptive and grid-aware charging decision-making in real time. The framework is evaluated under various EV penetration scenarios ranging from 30 to 100 EVs and is compared with uncontrolled charging, off-peak charging, and load-balancing strategies. The results demonstrate an average 25% reduction in peak load, a 15–20% reduction in charging costs, smoother feeder operation, and a user satisfaction rate exceeding 95% in meeting charging requirements. Furthermore, the proposed framework improves operational stability, reduces computational burden, and enhances charging coordination compared with conventional forecasting and heuristic scheduling approaches. These findings demonstrate the feasibility and scalability of integrating Machine Learning (ML)-based forecasting with real-time optimization for future smart-grid and EV energy management systems.