Optimized multi agent reinforcement learning algorithms with hybrid BiLSTM for cost efficient EV charging scheduling

With the fast development of electric vehicles, the demand for intelligent charging management strategies in order to minimize operational costs, ensure grid stability, and enhance user satisfaction. This paper proposes a new framework that embeds multi-MARL algorithm tuned by the Pelican optimization algorithm (POA) bidirectional long short-term memory for anticipatory energy forecasting scheduling in EV charging stations-EVCS. Unlike previous works that treat forecasting, the proposed method seamlessly unifies these steps, which were hitherto considered as separate entities: optimization and then scheduling. Components within a Markov decision process formulation. The framework employs publicly available Indian Energy Exchange (IEX) day-ahead market data, where POA-tuned BiLSTM forecasts electricity price and demand with improved accuracy, feeding into the MARL controller for dynamic scheduling. Experimental results demonstrate that the proposed method reduces charging cost by 12.34%, improves state-of-charge (SOC) satisfaction by 10.25%, and increases forecasting accuracy by 8.46% compared to conventional GA, PSO, MARL, and deep learning baselines. Furthermore, simulation time is reduced by 0.456 s, confirming computational efficiency. This study presents integrated frameworks that combine POA-tuned BiLSTM forecasting with a CTDE-based MARL architecture for anticipatory EV charging scheduling.