Refrigerant Charge Fault Diagnosis Of VRF System Based On EMD Denoising

As one of the main forms of HVAC systems,VRF system has the advantages of convenient installation,flexible design,intelligent control and stable operation.It has been widely used in various large buildings.With the promotion and popularization of VRF system,its energy conservation and safety issues will also receive widespread attention.Due to the complexity of working conditions in the actual operating state,VRF system will have various typical faults,which will affect its working performance,resulting in increased energy consumption,expensive maintenance costs,and even major safety accidents.Therefore,it is very necessary to timely detect and diagnose the common faults of VRF systems.This thesis uses data-driven method to conduct in-depth research on the failure of VRF refrigerant charge.According to the sample quantity and the corresponding refrigerant charge range,the original heating data of VRF refrigerant charge is divided into five types of fault levels,and the corresponding fault level labels are marked as L₁,L₂,L₃,L₄,L₅.Then,the box line graph analysis method is used to find out the outliers,and all the samples with the outliers are deleted.At the same time,the MIC algorithm and the Spearman correlation analysis method are used to select the final feature subsets.However,these data are directly collected from the measurement site and often contain a lot of noise,which will reduce the fault diagnosis rate of the model.Therefore,the EMD denoising is performed on the selected feature variables before final modeling to eliminate noise in the data and improve data quality.Finally,the denoised feature subset is used as the model input to establish a fault diagnosis model of VRF refrigerant charge,and the overall fault diagnosis rate,individual fault diagnosis rate and diagnostic time are used to evaluate the model.The results show that the feature selection combined with MIC algorithm and Spearman correlation analysis can effectively reduce the feature dimension,reduce variable redundancy and shorten the diagnosis time.Compared with the traditional BPNN model,the ELM model has better fault diagnosis performance,which improves the overall fault diagnosis rate of refrigerant charge from 79.83%to 83.88%,and the diagnostic time is reduced from 205s to 22s.The use of the EMD denoising algorithm further improves the performance of the model.Compared with the BPNN and ELM models before denoising,the overall fault diagnosis rates of composite diagnostic models of the EMD-BPNN and EMD-ELM are increased by 4.48%and 9.44%,respectively.Among them,the EMD-ELM model has the highest overall fault diagnosis rate,reaching 93.32%,and the diagnostic time is only 21s.In addition,the EMD-ELM model has a higher diagnostic rate for each fault level than the ELM model,with the fault level L₄ increasing the most,reaching 31.5%.In short,EMD denoising can indeed improve the quality of the data and contribute to the improvement of the fault diagnosis performance of the model.The fault diagnosis model of VRF refrigerant charge based on EMD denoising proposed in this thesis has certain research and application value.