Lithium-Ion Batteries state of health estimation based on optimized TCN-GRU-WNN
Nan Zhang, Jing Li, Yunfeng Ma, Kaichun Wu
Abstract
The exponential growth of the new energy vehicle sector has underscored the critical importance of precise state of health estimation for Lithium-Ion Batteries, which constitute the fundamental component of power systems in these vehicles. Addressing the limitations of conventional physical model-based estimation techniques, characterized by high computational demands and suboptimal real-time performance, this study introduces an innovative hybrid model. This model integrates a Genetic Algorithm (GA)-optimized Temporal Convolutional Network (TCN), Gated Recurrent Unit (GRU), and Wavelet Neural Network (WNN) to assess the state of health (SOH) of Lithium-Ion Batteries. Within this framework, the TCN extracts critical temporal features from battery operation data through temporal convolution operations, the GRU captures long-term dependencies of the data sequences, and the WNN performs multi-scale signal decomposition through wavelet transforms to enhance the model's nonlinear mapping capability. To maximize model efficacy, GA was utilized to conduct a comprehensive search for optimal parameter configurations. The results show that this hybrid method can estimate the SOH of lithium-ion batteries with high accuracy and robustness, and the average estimation error is significantly reduced to less than 1 %, thus offering valuable support for battery health management and contributing to enhanced safety and efficiency in both new energy vehicles and energy storage systems .