A hybrid deep learning and swarm intelligence framework for battery state of charge estimation and electric vehicle smart charging

Abstract

The increasing adoption of plug-in electric vehicles (PEVs) has heightened electricity demand, necessitating grid infrastructure upgrades. State of charge (SoC) is a very important parameter for batteries used in EVs. Accurate SoC estimation is pivotal to optimizing battery health, energy management, and driving range reliability in EVs. This study first introduces a Bidirectional Long Short-Term Memory with Loung Attention Mechanism (BiLSTM-LAM) model, achieving high SoC estimation accuracy with 0.73% mean absolute error (MAE), 1.23% root mean square error (RMSE), and a significantly reduced maximum absolute error (MAX) of 4.46%. Leveraging precise SoC predictions, charging and discharging power profiles are dynamically optimized to align with grid conditions, minimizing battery stress. Uncoordinated charging exacerbates power losses and distribution line overloads, increasing operational costs. This study then presents a coordinated charging strategy integrating charging and discharging operations to optimize PEV integration with distribution networks. An optimization model is formulated, balancing operational costs, renewable energy utilization, battery degradation cost, and grid constraints. The model is solved via particle swarm optimization (PSO), demonstrating 40.36–46.86% cost reduction across 10%–30% PEV penetration levels in a 33-bus radial network integrated with solar and wind energy. Results validate the strategy’s effectiveness in balancing grid stability with renewable intermittency.