摘要
为了提高多联机系统制冷剂充注量故障检测率,本文提出一种结合最大相关最小冗余(minimal-Redundancy-Maximal-Relevance,m RMR)和随机森林(RandomForest,RF)算法的在线故障诊断策略。利用mRMR结合RF在训练集上进行特征选择,结合网格搜索和十折交叉验证对随机森林、决策树和支持向量机3种模型进行参数寻优,并将优化后的模型用于测试集中。在多个在线验证集上对mRMR-RF模型进行验证,结果表明:仅选择6个特征变量就可建立准确率达98.63%的随机森林诊断模型;3种诊断模型中,RF算法的诊断效果最好,整体分类准确率达到97.06%;在4个不同在线多联机系统上检验,其分类准确率分别为95.82%、85.74%、88.24%和93.96%,分类准确率均在85%以上。说明基于mRMR-RF的故障诊断模型具有较强的泛化能力。
In order to improve the fault detection rate of variable refrigerant flow(VRF)system,an online fault diagnosis strategy based on minimal-Redundancy-Maximal-Relevance(mRMR)combining with random forest(RF)is proposed in this paper.The mRMR is combined with the RF for feature selection in training set.Based on grid search and ten-fold cross-validation,the parameters of random forest,decision tree,and support vector machine are optimized,and then the optimized model is tested on the test set.The mRMR-RF model is validated on multiple online verification sets.The results show that only 6 feature parameters can be selected to establish a random forest diagnosis model with an accuracy of 98.63%.Among the three diagnostic models,the diagnosis results of RF model are the best,and the overall classification accuracy rate is 97.06%.The model is tested in four different online multi-line systems with classification accuracy of 95.82%,85.74%,88.24%and 93.96%,respectively,and the classification accuracy rate is above 85%,so the generalization ability of fault diagnosis model based on mRMR-RF is strong.
作者
刘倩
李正飞
陈焕新
王誉舟
徐畅
LIU Qian;LI Zhengfei;CHEN Huanxin;WANG Yuzhou;XU Chang(School of Energy and Power Engineering,Huazhong University of Science and Technology,Wuhan,Hubei 430074,China)
出处
《制冷技术》
2019年第6期1-8,共8页
Chinese Journal of Refrigeration Technology
基金
国家自然科学基金（No.51876070,No.51576074）
关键词
多联机系统
故障诊断
最大相关最小冗余
随机森林
参数寻优
Variable refrigerant flow system
Fault diagnosis
Minimum Redundancy Maximum Correlation
Random forest
Parameter optimization
