AI-driven predictive energy storage management for electric vehicle battery systems using a hybrid PSO–RNN framework

Abstract

To enhance the performance, lifespan, and reliability of lithium-ion battery packs in Electric Vehicles (EVs), this paper proposes an adaptive predictive Battery Management System (BMS) based on a hybrid Particle Swarm Optimization–Recurrent Neural Network (PSO–RNN) framework. Unlike conventional BMS approaches that struggle with charge imbalance and inefficient energy utilization, the proposed system integrates predictive analytics with real-time optimization to proactively manage State-of-Charge (SoC) variations. The effectiveness of the proposed approach is validated using both real-world and simulated datasets. Experimental results demonstrate a 75% reduction in SoC deviation (from ±4.8% to ±1.2%), a 55.6% decrease in equalization time (from 45 min to 20 min), and a 67.7% reduction in energy loss (from 6.5% to 2.1%) compared to conventional BMS methods. Additionally, the system achieves a cycle life improvement of up to 23.7%, extending battery longevity significantly. The proposed model also improves prediction accuracy, achieving an R 2 value of 0.98, compared to 0.92 in traditional approaches. These results confirm that the PSO–RNN-based adaptive BMS provides superior charge balancing, enhanced energy efficiency, and extended battery life without significant additional hardware cost. The proposed framework offers a scalable and robust solution for next-generation EV battery systems and can be extended to Vehicle-to-Grid (V2G) applications under diverse operating conditions.