基于递归特征消除-加权k近邻算法的多联机系统制冷剂充注量故障诊断策略

Fault Diagnosis Strategy of Refrigerant Charge in Variable Refrigerant Flow System Based on Recursive Feature Elimination and Weighted k-Nearest Neighbor Algorithm

本文提出了一种基于递归特征消除-加权k近邻算法的多联机系统制冷剂充注量故障诊断策略。首先,基于专家先验知识筛选18个多联机运行特征,经数据预处理步骤后,采用递归特征消除(RFE)算法进行特征选择,筛选出最优特征子集;然后基于加权k近邻(wkNN)算法对训练集建立诊断模型,并采用网格搜索算法得到最优参数组合,对制冷剂充注量故障进行诊断。结果表明:该诊断策略弥补了现有方法中“只适用于单一工况、充注量等级分类少”等不足,选择重要性排列前7的特征集作为最优特征子集,在全工况和9个充注量等级的情况下,整体准确率为98.30%,受试者工作特征曲线下的面积(AUC)为0.9883,为设备维护人员提供了详细、关键的信息。

The fault diagnosis strategy of refrigerant charge in variable refrigerant flow(VRF)system is proposed in this paper based on recursive feature elimination(RFE)and weighted k-nearest neighbor(wkNN)algorithm.Firstly,based on the expert prior knowledge,18 VRF running features are screened.After the step of preprocessing data,the RFE algorithm is adopted to select the feature,and then the optimal feature subset is selected.Secondly,the diagnosis is established based on the wkNN algorithm,and the model and the grid search algorithm are used to obtain the optimal parameter combination to diagnose the refrigerant charge failure.The results show that the diagnosis strategy can make up for the shortcomings of“only applicable to single working condition and less classification of refrigerant charge amount level”in the existing method;the top 7 feature sets of importance ranking are selected as the optimal feature subset;the overall diagnostic accuracy is 98.30%and area under receiver operating characteristic curve(AUC)value is 0.9883 under the condition of overall working condition and 9 charging levels,which provides detailed and critical information for equipment maintenance personnel.