Research On Fault Diagnosis Of Chillers Based On KPCA-LSSVM

The energy consumption in buildings accounts for a large share of final energy use worldwide,which is still increasing every day.Most of the energy used in buildings is by the heating,ventilation,and air conditioning(HVAC)systems.In buildings,especially in large-scale buildings,water chillers are widely employed for air conditioning.However,operational faults in chillers will lead to a high energy consumption,a poor indoor thermal comfort and a low operational safety.Therefore,the operating faults should be detected and diagnosed timely and effectively.In this thesis,using the experimental data reported in ASHRAE RP-1043,a new fault detect and diagnosis(FDD)based on kernel principal component analysis(KPCA)and least squares support vector machine(LSSVM)for chillers is developed.The specific work is as follows:(1)Research on fault feature extraction method for chillers based on KPCAAiming at the serious correlation between the chiller variables and the high degree of nonlinearity,the multivariate statistical analysis method is used to extract the fault features.In view of the defect of the PCA method in nonlinear feature extraction,the KPCA method is introduced to extract high-order statistical information of system variables and reduce the data dimension,so that the sample data has better separability.The two methods are evaluated by the fault detection results,which indicates that the KPCA method is more suitable for feature extraction of chiller faults.(2)Research on a combined fault diagnosis approach for chillers based on KPCA-LSSVMIn view of the fact that the fault data is less in the actual operation of the chiller,and the relationship between faults and symptoms is complicated,it is difficult to diagnose faults.Therefore,LSSVM is introduced as a fault classification algorithm.Combining LSSVM with KPCA,a combined fault diagnosis approach based on KPCA-LSSVM is proposed.Firstly,KPCA is used to extract the nonlinear principal elements of the data,and then the extracted data is input into the LSSVM for fault classification.Through the diagnosis and analysis of seven common faults of ASHRAE RP-1043,and comparing with SVM,LSSVM and PCA-LSSVM methods,the results demonstrate that this method has higher accuracy,and also indicates the selection of two parameters of LSSVM for diagnosis has a large impact of the result,which needs to be further optimized.(3)Research on LSSVM fault diagnosis based on IGSA optimizationTo avoid the blindness of LSSVM parameter selection,an improved gravity search algorithm(IGSA)is proposed to optimize the parameters of LSSVM.The basic gravitational search algorithm(GSA)lacks memory.It is improved by introducing the speed update mechanism of particle swarm optimization(PSO),which increases the memory and information sharing ability of the algorithm.The performance test of the benchmark function shows that IGSA has faster convergence speed and higher search accuracy.Based on the KPCA feature extraction,the fault diagnosis model of IGSA-LSSVM is established and compared with the diagnosis model based on PSO-LSSVM and GSA-LSSVM.The results verify the effectiveness and superiority of the proposed method.