Fault Detection And Diagnosis Of Centrifugal Chillers Based On Machine Learning

Machine learning is one of the most important sub-directions in the field of artificial intelligence in the 21st century.With the increasing demand for intelligence in modern society,it is extremely urgent to promote the research and application of machine learning in some industrial fields with sufficient data.The popularity of machine learning is the first step in the concept of artificial intelligence.In the field of HVAC,centrifugal chillers are the most commonly used form of mainframe,which has complex structure and difficult maintenance.Therefore,it is worthwhile to apply the machine learning model on this issue.This paper describes the development history and research status of centrifugal chiller fault diagnosis technology at home and abroad,and compares the advantages and disadvantages of several major fault diagnosis methods.Based on the seven faults with the highest frequency of centrifugal chillers,analyze the faults.The coupling relationship between the parameters occurred,and the eight eigenvectors that best represent the chiller fault characteristics were selected to establish a neural network-based machine learning system.12557 sets of experimental data were extracted from ASHRAE RP-1043 and divided into training set,cv set and test set in a ratio of 6:2:2.The weights and thresholds of the BP neural network were obtained by fitting,the model was selected by the cross-validation set,and the diagnostic effect of the BP neural network was tested with the test set.Finally,the wavelet transform is used to denoise the input features,so that the neural network achieves higher precision.For the case of few fault samples in the actual operating conditions,this paper establishes an anomaly detection system based on multivariate Gaussian distribution,and uses the F1 index to combine the correct rate of classification to test the quality of the system.