Research On Hierarchical Hybrid Fault Type Recognition And Adaptive Refined Diagnosis Strategy Of VRF System

This dissertation establishes a hierarchical adaptive fault diagnosis mechanism for the occurrence of multiple types of faults that may occur in the VRF system,as well as the detailed level of occurrence or the cause of the fault.The mechanism first diagnoses the faulty objects for different faults,and independently selects the best refined diagnosis model according to the category of the fault to further diagnose the detailed fault category,which can reduce the negative influence caused by too many tags in one-step fault diagnosis and different faults using the same model for diagnosis.The hierarchical diagnosis mechanism can not only realize the diagnosis of specific faults with specific models,but also divide the severity level of the faults in more detail,which has greater value in popularization and application.Firstly,in order to simulate several failures and multiple severity levels that may occur in the actual operation of the variable refrigerant flow system,sufficient data for normal operation and multiple types of failures are obtained through experiments to build a hybrid fault hierarchical diagnosis mechanism.The mechanism is based on the Catboost algorithm to establish the first-level fault classification model,and uses the Prediction Values Change method in the Catboost package to rank the importance of features.Considering the model’s running time cost and diagnosis accuracy,10 characteristic parameters are selected in the final input model.Secondly,construct the detailed type discrimination model in the second layer of the diagnosis mechanism.In the article,two methods are adopted to reduce the data dimension.One is the association rules to find feature parameters that are strongly dependent on the fault label,and the other is the use of linear discriminant analysis for feature extraction.The purpose is to determine the different feature parameter subsets required for different faults,and to complete the second feature screening in this mechanism.At the same time,a refined diagnosis model for different types of faults is still established based on the Catboost model.Then,the fault detailed diagnosis effect of the feature parameter subset selected based on association rules and linear discriminant analysis is compared,and the dimensionality reduction method of association rules is finally selected.Finally,the test set is used to perform a complete test of the layered adaptive diagnosis mechanism to observe the diagnosis accuracy and robustness of the model.The results show that the overall accuracy of fault type recognition on the test data set reaches 99.29%.The test samples are isolated from each other according to the identified type and a matching detailed diagnosis model is selected autonomously.Finally,the probability of electronic expansion valve failure,liquid refrigerant floodback,and refrigerant charge failure to accurately identify the specific occurrence level or detailed type is 96.21%,99.67%,and 94.63%,respectively.The results show that the hierarchical diagnosis mechanism can better identify the fault categories and finely locate the fault occurrence level or cause,and has good generalization performance and promotion value.