Research On Soft Fault Diagnosis And Correction Method For Air-Conditioning System Based On Bayesian Inference

With rapid development of society,energy consumption of airconditioning system in China accounts for 40-50% of building energy consumption.Therefore,the method of reduce the energy consumption of air-conditioning system and develop energy-efficient technology has become a research hotspot.However,universal faults in air-conditioning system lead the poor performance of the energy-efficient technology and unnecessary energy waste.Current studies mainly focus on single type fault identification of air-conditioning system,but seldom consider multiple types faults at the same time.Additionally,fault level evaluation and fault self-correction are not further completed after fault identification.Thus,this dissertation proposed a novel fault diagnosis and self-correction method for air-conditioning system based on Bayesian inference,aiming at establishing an integrated mechanism of "fault state discrimination-fault source location-fault self-correction" for the air-conditioning system.Bayesian inference transforms the parameter evaluation problem into maximum likelihood function solving problem by defining distance function.First,a novel fault self-correction method for air-conditioning system based on physical mechanism model was proposed.The distance function of the proposed method was established by the physical mechanism among parameters,and the fault self-correction procedure does not require a large amount of data and supports multiple faults self-correction.In a detailed air-handing unit,the fault correction rate is not less than 99.20% in single fault scenarios and 92.23% in double fault scenarios.Secondly,a fault self-correction method of air-conditioning system based on Bayesian inference coupling with autoencoder was proposed.The autoencoder can extract the potential characteristic information of original input variables.After verification,the method has significant fault selfcorrection ability,and the sensor measurement error rate in single fault scenario and multiple fault scenario is reduced to 7.67% and 3.00%,respectively.Then,the influence of key factors on the Bayesian inference coupled autoencoder algorithm was investigated,and the performance of prior distribution updated and input dimension expansion strategies was verified.Finally,a fault detection and diagnosis method of air-conditioning system based on Bayesian inference was proposed.The distance function and fault diagnosis criterion were defined by the inherent relationship among variables.The results show that the system faults are accurately identified and the fault correction rate is as high as 98%.In addition,the proposed method still shows excellent fault diagnosis performance even with only a small number of datasets.