Anomaly Identification Model of Power Battery Self-discharge Based on Deep Belief Networks

Abstract

Electric vehicles will receive signal data such as voltage and current while the vehicle is running, which can reflect the safety status of the vehicle. Therefore, a quantitative description of the safety status of power batteries during operation can be achieved by analyzing and mining the safety features contained in the historical data of the vehicle that is running. However, considering the physical characteristics of the battery system and factors such as sensor design and acquisition accuracy, there is inevitably information coupling, redundancy, and error between different signals, making it difficult to extract and quantify safety features accurately. To solve this problem, this paper combines the identification method of abnormal risk characteristics of self-discharge of power battery, constructs abnormal risk characteristics of self-discharge of power battery, and completes the calculation and safety quantification of self-discharge risk characteristics. On this basis, by combining multiple safety features with the vehicle safety status obtained from the above analysis, a supervised learning sample data set is constructed, and the deep confidence network (DBN) is used to map and describe the vehicle safety feature data and the vehicle safety status, so as to achieve accurate judgment and fault identification of the vehicle comprehensive status.