Research On Optimizing Chiller Of Fault Detection And Diagnosis Based On Support Vector Machine

In the context of increasingly scarce energy,how to reduce energy consumption has become one of the current hot research topics in environmental protection.For example,in the field of daily building refrigeration,according to relevant data,the operating energy consumption of chillers of building refrigeration and air conditioning is an important part of the total energy consumption of buildings,accounting for about 40%.At the same time,the operating energy consumption of the chiller will increase significantly when the chiller fails.Therefore,it is particularly important to reduce the energy consumption of the chiller during faulty operation and to accurately diagnose the working status of the chiller in time.Based on the experimental data of ASHRAE RP-1043 and combining seven typical single-engine fault cases,this paper aims to find an efficient and stable fault detection method for centrifugal chillers,and proposes an improved Additional Weight One-Class Support Vector Machine(AWOCSVM)chiller fault detection model and the chiller fault diagnosis model based on Autoencoder combined with Support Vector Machine(AE-SVM),the specific work is as follows:(1)Since the signals collected during the operation of the unit are noisy,to solve the problem of outliers in the training samples of OCSVM,based on distance weight and local density weight is proposed.Check the model.According to the Euclidean distance between the training sample points,additional weights are assigned to each training sample.At the same time,principal component analysis(PCA)is used to reduce the dimensionality of the sample data to construct a single-class hyperplane for the normal operation of the chiller.A fault detection model based on D and T~2statistics is constructed.The experimental simulation results show that the fault detection model based on PCA-AWOCSVM can effectively increase the false alarm rate of 9%for the lowest level fault of SL1,and the false alarm rate of other faults can reach 0%.(2)Aiming at the characteristics of high dimensionality and high correlation of non-linear fault data of chillers.In order to preserve the high-dimensional characteristics of the original fault data to the greatest extent,the neural network dimensionality reduction method-autoencoder is introduced,and the fault classification model is referred to the support vector machine,and a new fault diagnosis model based on AE-SVM is proposed.Through the verification of seven typical single-shot fault data of the ASHRAE RP-1043 project,and the comparison of the two diagnostic models of PCA-SVM and KPCA-SVM,the experimental results show that this method has better diagnostic performance and an accuracy rate of up to 88.15%,which verified the effectiveness of the automatic encoder method.(3)Research based on the AE-SVM fault diagnosis model optimized for the Sparrow Search Algorithm.The selection of the―penalty parameters‖of the SVM classifier and the parameters of the RBF kernel function has an important impact on the diagnosis results.Therefore,in order to reduce the uncertainty of parameter selection,the SVM is searched for parameters by introducing a swarm intelligence optimization algorithm—Sparrow Search Algorithm.Comparing the parameter optimization results based on the GA-SVM and PSO-SVM methods,the AE-SVM model optimized by the sparrow search algorithm has improved accuracy of 6.23%,8.12%,5.95%and 1.87%for the four failure levels.