Fault Diagnosis Of Energy Efficiency Decline In Chiller Operation For Practical Application

The chiller needs to run for a long time in the cooling season,and the failure of the chiller will result in energy waste,unit life decline and insufficient cooling capacity,etc.Therefore,the study on fault detection and diagnosis methods of the chiller is getting more and more attention.However,the fault detection and diagnosis method for chiller has not been widely applied in practice due to the mismatch between the characteristics and field sensors and the noise of operation data.In this paper,from the perspective of practical application,the fault detection and diagnosis of the chiller are studied.First,in view of the feature selection stage,summarizes the domestic current situation of the computer room of the chiller sensor installation,and study a packaged feature selection method based on Particle Swarm Optimization(PSO),considering the false-detect and false-alarm caused the price of different,to introduce the Cost-Sensitive Classification Accuracy(CSCA),and combining the Support Vector Data Describtion(SVDD)classification model as the fitness function,find out the optimal feature subset and finally on the basis of the existing sensor types,determined in this paper,we study the characteristics of the collection.Secondly,for the data pre-processing stage,from the issue of filtering divergence as an entry point,the operation characteristics of the on-site chiller are analyzed.Aiming at the variable operation conditions and operation modes of chillers,based on Kalman filtering,the adaptive selection of fading memory(FM)and limited memory(LM)methods is realized,and LM is further improved.Furthermore,this method is fused with interactive multiple models(IMM).A non-linear KF method based on F/LM-IMM with variable memory length is proposed,which realizes the strong tracking characteristics and robustness of the filter when the chiller changes the operating conditions and operating modes.Finally,a real-time fault detection and diagnosis(FDD)model based on SVDD was established,so as to build a complete application-oriented chiller unit FDD system.Finally,an online FDD model based on SVDD is established,so as to build a practical application oriented FDD system for water chiller.In this paper,ASHARE RP-1043 experimental data are used to verify the effectiveness of the system.In the feature selection stage,the p SO-based feature selection method introduced in this paper determines the optimal feature subset containing 14 features,whose CSCA adaptive value reaches 0.9951,and then reduces the features to 9 according to the existing sensor features on the site.In the phase of data denoising,the data before and after filtering are first compared,and it is found that after the processing of F/LM-IMM nonlinear KF method based on variable memory length,the operation data of normal mode satisfying the steady-state judgment condition rises from 35% to 94%,greatly reducing the loss of operation data.Then,three authentication schemes are set up based on the three schemes to compare the filtering effect of the Unscented Kalman Filter(UKF),FM-UKF and the nonlinear KF method based on the variable memory length F/LM-IMM.The results show that the F/LM-IMM nonlinear KF method based on variable memory length is effective in data filtering and denoising of chiller under various operation conditions and modes.FDD phase,the real-time running data of the unit can be divided into the original data and the data after filtering denoising method,using SVDD model respectively for the two types of data verification,the results showed that: in fault detection phase,the sample after filtering denoising processing,the False Alarm Rate(FAR)and Correct Rate(CR)were improved.In the fault diagnosis stage,the operation data of the seven fault modes are denoised,and the diagnosis CR is significantly improved.The results show that the nonlinear KF method based on variable memory length F/LM-IMM has significant filtering effect and can improve the performance index of FDD.