Hybrid Chaotic Attractor Recurrent Network Transnet Architecture for Accurate State of Charge Estimation of Li-Ion Batteries in EV Application

Main objectives of the study are to design and validate a novel state of charge (SoC) estimation framework for Lithium-Ion Batteries (LIBs) in Electric Vehicle (EV) Energy Storage Systems (ESSs), integrating the chaotic attractor recurrent network (CARN) with transformer techniques. This hybrid approach aims to overcome limitations in conventional battery management systems (BMSs), particularly in handling noisy inputs, long-range dependencies, and data imbalance. These objectives were achieved by implementing a structured methodology that incorporates data balancing to mitigate skewed datasets, exploratory data analysis (EDA) for anomaly detection and pattern recognition, and feature scaling for input normalization, thereby ensuring robust and effective model training. The hybrid classification model leverages the temporal pattern recognition capability of ARN alongside the strong attention mechanism of the Transformer, enabling superior adaptability under diverse operating conditions. Implemented in Python, the proposed method was rigorously tested across multiple scenarios to confirm its reliability and accuracy. The most important results are the reduced root mean square error (RMSE) of 0.9671, mean square error (MSE) of 0.9352, mean absolute error (MAE) of 0.793, and an enhanced R²-score of 99.86%, which collectively demonstrate significant improvements over conventional estimation techniques. The significance of obtained results lies in validating the proposed model's ability to deliver highly accurate, robust, and real-time SoC prediction, thereby contributing to safer and more efficient battery management in EVs. This study highlights the potential of hybrid deep learning architectures to advance ESS safety, optimize energy utilization, and support sustainable electric mobility.