Research on State of Charge Estimation of Power Battery Based on Neural Network under the Background of New Energy Electric Vehicle

In order to study the state of charge (SOC) estimation method of lithium batteries for new energy electric vehicles, a method of applying deep learning to SOC estimation is proposed, which is based on the attention mechanism and temporal convolutional network (A-TCN), and extracts the key feature parameters from the power battery operation time series dataset for training and analysis and applies them to online estimation. In this study, voltage and current information are used as feature parameters, the attention mechanism weights the feature parameters, and the TCN performs SOC estimation of the current and past operating data of the power battery by means of the unique expansion convolution structure. In order to explore the estimation effect of A-TCN, a comparative analysis with the TCN training model is conducted. The training results indicate that: the prediction results of the model are affected by the training data, and the accuracy of the model varies when using different charge/discharge cycle data; the average absolute error of the designed model in the test reaches 0.56%, and the maximum error is 4.2%, which are both improved compared with the TCN training model, indicating that the A-TCN training model has a better estimation accuracy compared with the TCN training model The A-TCN based SOC prediction model maps the current and voltage data of the current battery operation to the corresponding SOC. Applying the A-TCN model to actual battery management can improve the accuracy and stability of the battery condition monitoring, discover the abnormal situation of the power battery in time, and take corresponding measures to avoid battery damage or accidents.