Chiller Fault Diagnosis Based On Canonical Variable Analysis

Heating Ventilation Air Conditioning(HVAC)system is not only an important part of the building,but also the highest energy consumption system.Related studies show that the energy consumption of chillers accounts for 40%-50% of the HVAC system.When a chiller fault occurs,it will not only affect the comfort level of the building,but also increase the energy consumption of the system.Therefore,it is an important link to ensure the safe and stable operation of the chiller and to save energy for timely detection and diagnosis of chiller faults.Based on ASHRAE RP-1043 project data set of American HVAC,the fault diagnosis method of chiller based on canonical variable analysis is studied in this thesis.Main works are as follows:(1)In order to solve the problem of the difficulty of incipient fault detection for chiller with dynamic characteristics,this thesis gives a method based on Improved Canonical Variable Analysis(ICVA).Firstly,Exponentially Weighted Moving Average(EWMA)method is taken to pre-process the operation parameters of the chiller.Then,take the processed data as the input of the canonical variable analysis method to extract the fault features and carry out incipient fault detection of the chiller.The experimental results show that this method can improve the detection rate of incipient faults of the chiller.(2)In order to solve the problems of obtaining the fault data and insufficient data samples in the actual operation of chiller uneasily,this thesis adopts the Least Squares Support Vector Machines(LSSVM)method in machine learning to diagnose the fault.The fault feature data extracted by ICVA method is taken into this thesis as the basis of fault diagnosis and establish the fault diagnosis model of ICVA-LSSVM.Meanwhile,The experimental results show that the proposed method has higher accuracy of fault diagnosis.(3)To solve the problem of parameter selection in ICVA-LSSVM model,this thesis gives adaptive differential evolution particle swarm optimization algorithm.In this method,the sine formula of the scaling factor is introduced,and then uses the velocity and position updating formula of particle swarm optimization to construct a new mutation operation and improve the global and local search ability to accelerate the convergence speed.This thesis adopts the adaptive differential evolution particle swarm optimization method to optimize the parameters of the LSSVM model.Meanwhile.The results show that the ADEPSO-LSSVM model has higher fault diagnosis performance.