Improved back-propagation neural network-multi-information gain optimization Kalman filter method for high-precision estimation of state-of-energy in lithium-ion batteries

The accurate estimation of the state-of-energy (SOE) is crucial for extending battery life and improving the performance of electric vehicles. To address the issue of low estimation accuracy of SOE in lithium-ion batteries, a back-propagation neural network-multi-information gain optimization Kalman filter (BP-MIGKF) method is proposed. First, an adaptive mechanism based on dual-indicator feedback is used to improve the forgetting factor recursive least squares algorithm, enhancing the robustness of parameter identification for the second-order RC-PNGV equivalent circuit model. Second, the BP neural network is employed as the architectural framework for SOE estimation. Specifically, the training set includes current, voltage and all model parameters to efficiently capture the complex nonlinear relationships within the battery. Then, the MIGKF algorithm is used to filter noise from the SOE estimation values. Additionally, by predicting the error information generated at each iteration step and combining different weighting factors, the cumulative error of the terminal voltage is corrected and suppressed. Furthermore, the Kalman gain factor selected based on targeted experiments is used to optimize the gain matrix, further enhancing the accuracy of SOE estimation. Finally, several dynamic driving experiments at different temperatures are designed and executed to validate the performance of the proposed method. The experimental results indicate that the SOE estimation error based on the proposed method is controlled within 3 % under the premise of rapid convergence and strong robustness. This paper provides an effective way to combine data-driven and classical filtering for battery state estimation.