基于BPNN的地铁空调制冷剂充注量故障诊断权重分析

Weight analysis of train air conditioning system based on BPNN

基于数据记录、充注量正常标签及不同级别的制冷剂泄漏故障试验建立故障诊断模型。对试验数据进行清洗后,将数据的80%与20%分为训练集与测试集,利用训练集的数据训练反向传播神经网络(BPNN)模型,并对模型进行测试得到混淆矩阵结果,测出模型的准确率,分析发现BP神经网络(BPNN)的故障诊断预测总准确率达到了99.609%。为减少特征变量的复杂程度,简化机器学习的复杂度,对原有的36个特征变量进行重要程度的分析。设置了不同神经元个数的BPNN,利用权重值矩阵分析特征变量在神经网络传播过程中对诊断结果的影响程度。

A fault diagnosis model is established based on data recording,normal charge label and different levels of refrigerant leakage fault experiments.After cleaning the experimental data,80%and 20%of the data are divided into training set and test set.The back propagation neural network(BPNN)model is trained with the data of the training set,and the confusion matrix results are obtained by testing the model.The accuracy of the model is measured.It is found that the total accuracy of fault diagnosis and prediction of BP neural network reaches 99.609%.In order to reduce the complexity of feature variables and simplify the complexity of machine learning,this study analyzes the importance of the original 36 feature variables.The BP neural network with different number of neurons is set up,and the influence of characteristic variables on the diagnosis results in the process of neural network propagation is analyzed by using the weight value matrix.