State of health estimation for lithium-ion batteries based on short-term and global dependency information

Accurate knowledge of state of health (SOH) is essential to enhance operational safety and reliability of lithium-ion batteries, and machine learning is widely employed based on partial capacity variation, such as incremental capacity (IC) analysis. However, those algorithms often encounter peak disappearance, data distortion, and difficulty in capturing long-term and short-term dependencies. To address these challenges, an efficient SOH estimation method is proposed based on short-term and global dependency information. Firstly, the maximum tangent angle and corresponding time within the peak range of IC curve are extracted as health features. Secondly, a parallel network integrating Informer network and long short-term memory network is constructed to map the relationship between health features and SOH, thereby effectively capturing both local and global degradation information. Thirdly, a self-attention mechanism is developed to dynamically assign weights to the outputs of networks, further enhancing feature extraction and information representation capabilities. The Bayesian optimization algorithm is developed to determine the hyperparameters of the network. Experimental validations are conducted on different cells and various types of batteries, and the results show that the proposed method can accurately estimate SOH with an error of 2 %, offering higher accuracy and robustness compared to traditional time-series forecasting methods.