基于优化熵算法的真实场景下电动汽车动力电池系统故障诊断  

Fault Diagnosis of Power Battery System for Electric Vehicles Based on Optimized Entropy Algorithm in Real Scenarios

作　　者：刘启全 马建[1] 赵轩[1] 张凯[1] 孟德安 LIU Qi-quan;MA Jian;ZHAO Xuan;ZHANG Kai;MENG De-an(School of Automobile,Chang'an University,Xi'an 710064,Shaanxi,China)

机构地区：[1]长安大学汽车学院,陕西西安710064

出　　处：《中国公路学报》2024年第10期233-248,共16页China Journal of Highway and Transport

基　　金：国家自然科学基金项目(52172362);陕西省重大科技专项项目(2020ZDZX06-01-01);霍英东青年教师基金项目(171103);陕西省科技成果转化项目(2024CG-CGZH-19);陕西省重点研发计划项目(2020ZDLGY16-01,2020ZDLGY16-02,2021ZDLGY12-01)。

摘　　要：动力电池系统故障诊断是保障电动汽车安全可靠运行的关键,其中避免误报警不仅能减少驾驶人对车辆的安全焦虑,同时也是诊断方法进行实际应用的必要条件,因此提升方法的可靠性具有重要意义。动力电池系统中电压的异常波动是电池性能恶化所释放的重要信号,因此能很好评估数据离散程度的熵方法在电池故障诊断领域被广泛研究。然而,当经典的基于区间概率的香农熵方法在进行工程实践验证时,发现结果中存在大量的一级和二级误报警单体电池。基于此,为了提升方法的准确率,首先,利用具有热失控事故的车辆电压数据分析模型的故障诊断原理。进一步地,基于正常车辆运行数据,研究2种典型场景下模型的误报警机制。在上述条件下,提出2种措施来缓解原始方法的误报和漏报问题,即数据优化法和核密度估计与熵融合法。最后,选取具有不同故障特征的真实故障样本来检验算法的泛化能力,同时他们的有效性和可靠性被分别验证。基于大量正常在役车辆数据,进行了模型优化前后性能的对比分析。结果显示:相比之下,2种方法对正常车辆的相对误报警率分别降低了90%和98%,从而极大提升了诊断策略的可靠性,推动了方法的在线实车应用,并为其他故障诊断策略准确率的分析和优化提供了思路。Fault diagnosis of power battery systems is key to ensuring the safe and reliable operation of electric vehicles,in which the avoidance of false alarms not only reduces the driver's anxiety regarding vehicle safety but is also necessary for the practical application of the diagnostic method.Therefore,it is crucial to improve the reliability of the method.Abnormal voltage fluctuations in a power battery system are critical signals released by the deterioration of battery performance;hence,entropy methods,which can satisfactorily assess the degree of data dispersion,have been widely studied in battery fault diagnosis.However,when the classical Shannon entropy method based on interval probability was validated in engineering practice,many primary and secondary false-alarm single cells were found in the results.Vehicle voltage data with thermal runaway accidents were first used to analyze the fault diagnosis principle of the model to improve the accuracy of the method.Furthermore,based on normal vehicle operation data,the false-alarm mechanisms of the model in two typical scenarios were investigated.Under the above conditions,two measures were proposed to mitigate the false and missing alarm problems of the original method:the data optimization method and the kernel density estimation and entropy fusion method.Finally,real fault samples with different fault characteristics were selected to test the generalization ability of the algorithms,and their validity and reliability were verified separately.Based on a large amount of normal in-service vehicle data,a comparative analysis of the performance before and after model optimization was conducted.The results show that the relative false alarm rates of the two methods on normal vehicles decrease by 90%and 98%,respectively.Thus,this study significantly improves the reliability of the diagnostic strategy,promotes the online real-vehicle application of the methods,and provides ideas for analyzing and optimizing the accuracy of other fault diagnostic strategies.

关 键 词：汽车工程 故障诊断 香农熵 动力电池 数据优化 核密度估计 

分 类 号：U469.72[机械工程—车辆工程]