The Savitzky-Golay filter based bidirectional long short-term memory network for SOC estimation

This paper investigates a Savitzky-Golay filter based bidirectional long short-term memory network (SG-BiLSTM) by using the Adam algorithm for the state of charge (SOC) estimation of lithium batteries. In this hybrid method, a BiLSTM network is constructed to estimate SOC by using the discharge current and the terminal voltage as inputs, the Adam algorithm is adopted to update the weights and biases of the BiLSTM, and the SG filter is introduced to process the estimated SOCs. In the experimental part, the urban dynamometer driving schedule (UDDS) profile is performed on a battery test platform for data acquisition. In the simulation part, the root mean squared error (RMSE) and the coefficient of determination (R-2) is used to evaluate the model performance under different cases. The estimation results indicate that: the SG-BiLSTM has faster convergence speed and higher estimation accuracy when compared with other methods; the SG-BiLSTM shows strong robustness when applied to the data set with random noises added; appropriately increasing the hidden neurons helps to improve the model performance, but excessive increase will lead to overfitting.