Study On Refrigerant Charge Fault Detection And Diagnosis Of Variable Refrigerant Flow System On The Basis Of Support Vector Machine

Fault detection and diagnosis(FDD)has actual significance for improving stability,economy and safety of refrigeration and air conditioning system.Data-based FDD methods have been wildly used in many domains for the past few years.This paper conducts a research on refrigerant charge FDD of variable refrigerant flow(VRF)system.The refrigerant charge fault developing mechanism is analyzed and the experimental system is described firstly.In addition,the structure of the original data and the FDD flow are introduced in detail.Correlation analysis(CA)and importance of variables are employed to select feature variables.Results show that 11 variables choose from the total 18 variables in original data are used to train and test model.Then the support vector machine(SVM)algorithm is applied to establish the FDD model.Based on the default settings of SVM parameters(i.e.C = 1,gamma = 0.09),the FDD accuracies of training data and testing data are 88.17%,87.79%,respectively.In order to further optimize the SVM model and improve the accuracies,parameters optimization are of vital importance.Grid search(GS)and 10-fold cross validation(CV)are used to find the optimal values of C and gamma.Results indicate that the optimal parameters are C = 1000,gamma = 0.25.Next,the best SVM model is built by the optimal parameters settings.Finally,the FDD accuracies of the best SVM model is 99.75% for training data and 99.05% for testing data.Obviously,the accuracies improve more than 10% with optimized parameters.Moreover,the results also validate that the effectiveness of parameter optimization methods.For purpose of improving the applicability of FDD model and realizing on-line running,a graphic user interface is developed by C# program.There are two child interfaces: a data preview interface and a SVM modeling interface.In the data preview interface,it can import the data from a CSV file,display the original data and show the essential information of variables.In the SVM modeling interface,it can train and test the SVM model when the necessary parameters are set.When finish training or testing the SVM model,the FDD accuracies and corresponding confusion matrixes will be displayed in the panel.Besides,the results can be saved to the local disk.