Lithium-Ion Battery State of Health Estimation Method Based on Frequency-Domain-Assisted Data Decoupling and Time–Frequency Features Compensation Learning Network

Accurate state of health (SOH) estimation is critical for lithium-ion battery maintenance and safety. Currently, numerous time series models based on deep learning have been applied to online prediction of SOH. However, the strong nonlinear coupling between different sample data seriously interferes with the estimation accuracy of the existing models. Rational network expansion can alleviate the issue, but limited samples pose a significant challenge to model training. Therefore, how to effectively reduce the strong nonlinear coupling between data is the key to achieve high accuracy prediction of SOH. In view of this, this article proposes an SOH estimation method based on frequency-domain-assisted data decoupling and time-frequency features compensation learning network (FA2D-TFCLNet). Specifically, this article proposes a Mel filtering algorithm to achieve preliminary nonlinear decoupling of the discharge data. Subsequently, time-domain features and frequency-domain features are fed in parallel to TFCLNet, which consists of a parallel multiscale convolutional attention (CPMCA)-based local receptor and Transformer. By organically combining the refined local aging features mined by the CPMCA and the discriminative global semantic relations captured by the Transformer, the proposed TFCLNet effectively realizes the comprehensive perception and deep decoupling of discharge data, thereby obtaining competitive SOH estimation accuracy. To validate the effectiveness of the proposed method, relevant experiments are conducted on NASA PCoE and Oxford datasets, and the results prove that the method has a high accuracy in estimating battery SOH.