An SOH Estimation Method Based on Knee Point Detection and Temporal Transfer Learning for the Lithium-Ion Battery Considering False Data Injection

With the widespread deployment of large-scale energy storage systems, this paper proposes a state of health (SOH) estimation method for lithium-ion batteries operating under fluctuating renewable power conditions. First, a random forest algorithm is employed to eliminate false data from the raw dataset. Then, the maximum mean discrepancy method is introduced to efficiently identify aging domains and detect knee points, thereby reducing the distribution discrepancy between the source and aging domains and enhancing model robustness. Subsequently, a wrapper-based feature selection approach is utilized to extract highly correlated features and establish mapping relationships with the prediction model. Finally, a hybrid model integrating the horned lizard optimization algorithm, bidirectional temporal convolutional network, and bidirectional gated recurrent unit is developed for SOH prediction. Experimental results demonstrate that the proposed method achieves mean absolute percentage error and root mean square error within 0.6% and 1.4% on the Oxford dataset, respectively, and exhibits strong performance on the National Aeronautics and Space Administration dataset, confirming its high accuracy and robustness.