Voltage-temperature aware thermal runaway alarming framework for electric vehicles via deep learning with attention mechanism in time-frequency domain

Timely and reliable thermal runaway alarming method for power battery pack plays a vital role in ensuring safe operation of electric vehicles. However, current methods neglect the coupling properties of battery data in time-frequency domain and rely on only one variable, namely temperature or voltage, to design alarming scheme, which is not sufficient to realize robust alarming. To overcome above problems, this paper proposes a novel voltage-temperature aware thermal runaway alarming approach using advanced deep learning model. The method has three main innovations. Firstly, wavelet analysis is used to extract frequency features from time -series data to reveal time-frequency coupling properties. Secondly, deep learning with attention mechanism is adopted to map the time-frequency representation of history data to predicted data. Thirdly, voltage-temperature joint alarming is proposed to improve diagnosis precision and robustness. Experiments show that the method has only 0.28% combined relative error for temperature and voltage prediction in a 7min time window and can achieve 8-13 min ahead thermal runaway prediction in real-world scenarios.

Automated summary

This paper proposes a voltage-temperature aware thermal runaway alarming approach using advanced deep learning model, which improves the accuracy and robustness of the alarming system. Wavelet analysis is used to extract time-frequency features, deep learning with attention mechanism is adopted to map historical data to predicted data, and a voltage-temperature joint alarming method is proposed. Experiments show that the method has a combined relative error of only 0.28% for temperature and voltage prediction in a 7-minute time window and can achieve 8-13 minute ahead thermal runaway prediction in real-world scenarios.