A Thermal Runaway Early Warning Method for Electric Vehicles Based on Hybrid Neural Network Model

Abstract

New energy vehicle has gradually become a new trend in global transportation development due to the renewable and environmentally friendly fuel they consume. At the same time, the charging safety issue of lithium-ion batteries for the electric vehicle limits the development of the industry. From the perspective of the electric vehicle charging data and based on the timing characteristics that lithium-ion battery charging has, this paper proposes a hybrid neural network electric vehicle thermal runaway temperature warning model, which combines an attention mechanism (AT), a temporal convolutional network (TCN), and a long- and short-term memory network (LSTM). Firstly, the charging temperature of the electric vehicle is predicted by establishing hybrid neural networks model, then comparing the real-time charging data with the predicted data, calculating the residual difference between the two. Analyzing the residual difference by using the sliding window method and then calculating the pre-warning threshold. Finally, realizing the thermal out-of-control early warning based on the residual difference to complete the monitoring of the charging status of the electric vehicle. The experimental results show that the AT-TCN-LSTM charging early warning model has higher accuracy and faster speed than other models, so that the method can accurately and quickly respond to charging accidents and achieve the early warning effect. At the same time, in order to verify the generalization ability of the model, transfer learning is used to transfer the trained AT-TCN-LSTM model to the charging data of different charging times and different vehicle types, and the results show that the model after transfer learning still has more accurate prediction accuracy.