Abstract:

With the development of the electric vehicle industry in China， the safety performance of the EV batteries has became one of the most important guidelines scaling the overall product value， whereas the detection of the battery thermal runaway is the most concerned safety issue of passengers. To solve the problem of the low accuracy of the conventional thermal runaway detection method， this paper extracts the state feature vectors from the voltage， current， time and other parameters directly observed by the battery sensors and optimizes and screen the features using the Gaussian mixed model. The mixed probability distribution of different safety states of the power battery is evaluated and the hidden Markov method is established by the BW algorithm. The current battery health status is judged by calculating the similar probability of the current observation sequence using the Viterbi algorithm. The experiment results show that the hidden Markov model can more accurately predict the thermal runaway status in real-time compared with the traditional time series detection method. It can realize preliminary thermal runaway risk detection without additional electrochemical testing experiment on the batteries， which improves the safety detection efficiency and reduces the detection cost.

Key words: battery thermal runaway, real-time alert, hidden Markov model, Gaussian mixed model, machine learning