Accurate state of health estimation using hybrid algorithm for electric vehicle battery pack performance and efficiency enhancement

Abstract

Temperature fluctuations, overcharging, and over-discharging are all issues that can cause fast deterioration, capacity loss, and thermal runaway in lithium-ion batteries (LIBs). To overcome these challenges, a hybrid model combining a stacked recurrent neural network (SRNN) and bidirectional long short-term memory (biLSTM) is presented for a reliable state of health (SoH) estimate. This model finds subtle patterns in battery data using SRNN layers to capture sequential dependencies and biLSTM modules to solve long-term temporal correlations while avoiding vanishing gradient concerns. The effectiveness of model is assessed by performance measures such as root mean square error (RMSE), mean absolute error (MAE), and maximum error (MAX), which demonstrate its superiority for precise SoH estimation. The stacked RNN-based SoH estimation achieves superior accuracy, with RMSE, MAE, and MAX error levels of 1.5%, 0.8%, and 4.84%, respectively, compared to GRU’s higher errors (3.8%, 3%, and 5.5%). Stacked RNN hierarchically processes sequential battery data, effectively capturing complex temporal relationships, and ensuring accurate and reliable SoH estimation for LIBs.