State of charge estimation of lithium-ion battery using denoising autoencoder and gated recurrent unit recurrent neural network

To reduce the influence of the measurement data noise on state of charge (SOC) estimation, a novel neural network method is proposed by combining an input data processing method with the conven-tional gated recurrent unit recurrent neural network (GRU-RNN) method. First, a denoising autoencoder neural network (DAE-NN) is introduced to extract the useful data features by reducing the noise and increasing the dimensions of the battery measurement data. Then, the feature-extracted data is utilized to train the GRU-RNN, which is widely used in SOC estimation. Notice that a good input data processing method plays a key role in the SOC estimation performance and the generalization ability. Therefore, it is not trivial to combine the data processing method (DAE-NN), and the SOC estimation method (GRU-RNN), which is named DAE-GRU. Compared with the traditional GRU-RNN, the new DAE-GRU method shows a better nonlinear mapping relation between the measurement data and the SOC because of the involvement of the DAE-NN. Finally, three different driving cycles are given in the experiment to cross -train and verify the proposed DAE-GRU, GRU-RNN and RNN. Compared with the GRU-RNN and the RNN, it is demonstrated that the proposed DAE-GRU has better accuracy and robustness in the SOC estimation. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

A novel method combining a denoising autoencoder neural network with a gated recurrent unit recurrent neural network is proposed to reduce the impact of measurement data noise on state of charge estimation. Experimental results demonstrate that the proposed method has better accuracy and robustness in SOC estimation.