An Adaptive LSTM Network With Fractional-Order Memory Unit Optimized by Hausdorff Difference for SOC Estimation of Lithium-Ion Batteries

This brief proposes an adaptive LSTM network with fractional-order memory unit optimized by Hausdorff difference to enhance the accuracy of state of charge (SOC) estimation. First, the Hausdorff difference is introduced to improve the memory unit in the LSTM network, and the order is introduced to enhance the degree of freedom of the network. Then, an adaptive adjustment approach of the order is used to dynamically regulate the ratio of the input information and forgotten information, and the adaptive LSTM network with the fractional-order memory unit optimized by Hausdorff difference (LSTM-H) is utilized to enhance the accuracy of SOC estimation. The performance of the proposed method and the LSTM network for SOC estimation is compared. The experiments verify that the LSTM-H network achieves the satisfactory SOC estimation results for different operating conditions.