电动汽车动力电池电压异常故障诊断方法

A Diagnosis Method for Abnormal Battery Voltage of Electric Vehicles

随着环境污染和能源危机越来越严重,以电动汽车为主的新能源汽车已成为未来车辆发展的主要趋势和焦点。在车辆运行过程中及时、准确地预测电压异常是预防电池发生故障的关键。通过高采样频率采集了20台电动汽车的车辆运行数据,并对获取的车辆数据进行数据清洗与处理。通过分析研究高频采样率下的车辆运行数据与电池电压间的关系,提取挖掘了与电池单体电压相关的特征指标,分别为:电池剩余电量(SOC)、加速踏板行程值、总电流、车速。利用长短时记忆神经网络(LSTM)和平均差异模型(MDM)理论,建立了电压异常故障诊断模型。根据模型输出结果,计算电池平均电压预测值和电池单体电压观测值的差值,运用统计学方法,验证了单体电压差值服从正态分布,取99%置信区间的临界值±0.06 V作为电池单体电压异常的一级故障报警阈值,二级和三级报警阈值分别为±0.12 V,±0.18 V。以故障车中发生的22次电压异常报警为测试数据,基于该模型对实际发生的电压异常故障进行诊断,共诊断出20次故障,准确率为90.91%,验证了阈值设定的合理性。研究实现了故障单体精准定位,并且能至少在150 s前检测出故障电池单体电压的异常变化,证明模型具有实时性、有效性和准确性,可为道路交通安全管控策略的制定、电动汽车实时故障诊断和故障定位提供理论支撑。

With the increasingly serious environmental pollution and energy crisis,new energy vehicles,mainly electric vehicles,have become the main trend and a focal point of vehicle development.Timely and accurately predicting abnormal battery voltage during vehicle operation is the key to preventing battery faults.High-sampling-rate operation data of 20 electric vehicles are collected,cleaned,and processed.By studying the relationship between the high-sampling-rate operation data and the battery voltage of these vehicles,several indicators related to the cell voltage are identified,including state of charge(SOC),stroke value of accelerator pedal,total current,and vehicle speed.Based on Long Short Term Memory(LSTM)network,combined with the mean difference method(MDM),a diagnosis model of abnormal battery voltage is developed.According to the output of the model,the difference between the predicted average battery voltage and the actual battery cell voltage is calculated,and a statistical method is used to verify that the difference of cell voltage follows a normal distribution.The critical value±0.06V of the99%confidence interval is taken as the first-level threshold of fault alarm for abnormal battery cell voltage,and the second and third-level alarm thresholds are set at±0.12V,±0.18V,respectively.Based on the test data of 22 abnormal voltage faults of electric vehicles,a total of 20 faults are detected based on the proposed method,with an accuracy of 90.91%,which verifies the effectiveness of the threshold setting.Moreover,the proposed model can precisely locate the faulty cells of abnormal battery voltage at least 150 s before it occurs,which shows that the model has good performance for real-time,valid,and accurate diagnosis of abnormal voltage of electric vehicles.It can provide theoretical support for safety strategy,real-time fault diagnosis and fault location of electric vehicles.