基于改进CNN和信息融合的动力电池组故障诊断方法

Fault Diagnosis of Traction Battery Pack Based on Improved Convolution Neural Network and Information Fusion

本文中提出一种基于改进CNN和信息融合的锂离子电池组故障诊断方法。首先,在全连接层前加入Concat层,以融合不同层次的特征,建立改进的CNN模型。然后,采用MCE准则来优化交叉熵损失函数,解决其在非标签维梯度不做处理的问题。针对人工确定CNN卷积核结构存在的不够实用的问题,利用BIC准则确定模型最优卷积核结构。采用以上改进CNN进行诊断后,引入确诊条件进行判断。对不符合确诊条件的诊断结果,进一步采用一般CNN网络进行辅助诊断。通过将初步诊断结果和辅助诊断结果采用D-S证据理论进行融合的方法,得到最终的诊断结果。测试结果表明,本文中提出的方法可有效提高动力电池组的故障诊断准确率。

A fault diagnosis method for lithium-ion battery based on improved CNN and information fusion is proposed in this paper.Firstly,the Concat layer is added before the fully connected layer to fuse the features of different levels,so an improved CNN model is established.Then,the MCE criterion is used to optimize the cross entropy loss function,solving the problem that the non-label dimension gradient is not processed.In view of the insufficient practicality of the convolution kernel structure manually determined,the BIC criterion is used to determine the optimal structure of convolution kernel of the model.After diagnosis by using the improved CNN mentioned above,the conditions for confirmed diagnosis are introduced for judgment.For the diagnosis results not meeting the conditions,the general CNN network is further used for auxiliary diagnosis.By fusing preliminary diagnosis result and auxiliary diagnosis result using D-S evidence theory,the results of final diagnosis are obtained.The test results show that the method proposed can effectively enhance the accuracy of fault diagnosis on traction battery packs.