CVR Method: Comprehensive Evaluation of Li-ion Battery State of Health with Residual LSTM

Lithium-ion batteries are widely used in industries such as automotive and electronics. Due to safety concerns, State of Health (SOH) evaluation is of great importance for users. To ensure safety, this paper proposes a novel approach named CVR Method for SOH online evaluation. In order to comprehensively evaluates the health states of the battery, CVR Method selects four easily obtainable health indicators that represent three aspects of battery performance: capacity, internal resistance, and open-circuit voltage. And, to reflect the real health status of batteries and alert users, the indicator with the poorest performance is chosen to represent SOH. CVR Method utilizes Local Outlier Factor (LOF) metric and Lagrange interpolation to handle anomalous data, employs linear normalization to standardize dimension. Then applies an optimized neural network model that combines LSTM with residual block for SOH prediction and adopts Shapley values to quantitatively analyze the impact of the four health indicators on SOH. Experimental results demonstrate that this evaluation approach aligns with the actual usage scenarios of the battery and reports a high accuracy of RMSE 1.46%.