基于大数据驱动的集成模型车辆热失控预测

Prediction of vehicle thermal runaway based on data driven ensemble learning model

电池故障是新能源汽车热失控的主要威胁之一,开发一种算法预测汽车电池是否以及何时发生热失控,以便及时发送高温预警信息成为迫切需要。热失控的原因复杂而又多面,热失控或触发于动力电池内部,也可由外力触发,通过单纯的物理模型做出精确的预测较为困难。因此,构建了一种集成的机器学习算法,通过分别考虑电压和温度、异常电流、单电池一致性和过充电风险因素,构建集成模型。该集成模型由五个子模型组成,这些子模型与通过网格搜索选择的超参数相关联。为实现更加准确的预测,该方法基于实测大数据进行训练,算法实用且灵活,可预测现实场景中锂电池热失控的可能性。实验结果表明:综合误报率为0.1656,验证了该方法的可行性。

Battery failure is one of the main threats to the thermal runaway of new energy vehicles.It is urgent to develop an algorithm to predict whether and when the battery is out of control,so as to send high temperature warning information in time.The causes of thermal runaway are complex and multifaceted.The thermal runaway or triggered to the power battery can also be triggered by external forces.It is difficult to make accurate prediction by simple physical model.Therefore,an integrated machine learning algorithm was constructed,and an integrated model by considering voltage and temperature,abnormal current,single battery consistency and overcharge risk factors was constructed respectively.The integrated model consists of five sub models,which are associated with the super parameters selected by grid search.In order to achieve more accurate prediction,the method was trained based on the measured big data.The algorithm was practical and flexible,and could predict the possibility of lithium battery thermal runaway in real scenes.The experimental results show that the comprehensive false alarm rate is 0.1656,which verifies the feasibility of the method.