Effect of data enhancement on state-of-charge estimation of lithium-ion battery based on deep learning methods

State-of-charge (SOC) estimation is the key to the safe and efficient utilization of lithium-ion batteries. With the development of deep learning method, SOC estimation methods based on neural networks are widely utilized. However, it has strict requirements on the data used with the neural network-based SOC method, and the widespread noise in the data would cover up the original characteristics of the data. Preprocessing the data to denoise and enhance the data may be able to improve the accuracy of the SOC estimation model, but relevant studies are still lacking. In this paper, average-step method, gaussian noise injection method, discrete wavelet transform method and empirical mode decomposition method are adopted to preprocess the data in order to study the effect of data enhancement on SOC estimation. All the methods are evaluated through a convolutional neural network-long short-term memory (CNN-LSTM) model. The improvements of root mean square error (RMSE), mean absolute error (MAE) and maximum absolute error (MAXE) in total dataset are 40.16 %, 38.00 % and 54.02 % using the average-step method, 3.94 %, 2.00 % and 4.98 % using the gaussian noise injection method, 16.54 %, 18.00 % and 21.87 % using the discrete wavelet transform-average-step method, 13.39 %, 11.00 % and 23.31 % using the empirical mode decomposition-average-step method.