Advances in Lithium-Ion Battery SOC Prediction: A Data-Driven GRU-RNN Framework

Abstract

This study examines an enhanced approach for estimating the State of Charge (SOC) for Lithium-ion batteries, employing a Gated Recurrent Unit (GRU) incorporated into a Recurrent Neural Network (RNN). GRU-RNNs are recommended over conventional techniques for their skill in handling time-series data, overcoming long-range dependence difficulties that confront classical RNN s. The proposed model incorporates a complex four-layered design, including a GRU hidden layer of 500 neurons. Optimization is accomplished utilizing the Adam Optimizer throughout 50 epochs. the Panasonic 18650PF dataset was exploited, comprising a range of drive cycles at several temperatures, to train and test the model. Performance tests, done using Mean Absolute Error (MAE) and Maximum Error (MAX), illustrate the model's accuracy in SOC estimate. The results show that GRU-RNNs provide a viable solution for the real-time, dynamic prediction of SOC in electric car battery management systems, possibly stimulating future developments in sustainable transportation technology.