Research On Fault Detection And Diagnosis Of Refrigeration System Based On LOF-RF

With the development of industrial technology and the improvement of people’s living standards,refrigeration equipment and technology are widely used in all walks of life and millions of households.They account for an increasing proportion of social energy consumption,accounting for about 15%of total social energy consumption.The issue of energy efficiency of refrigeration equipment has become a hot spot of social concern.With the extensive use of refrigeration equipment,the probability of failure will increase,and the operation of equipment with failure will inevitably lead to a significant decrease in energy efficiency.How to quickly and accurately complete the fault detection and diagnosis of the refrigeration system is extremely important.This article aims to seek an efficient and stable fault detection and diagnosis method,and to carry out research around the refrigeration system.Aiming at its typical faults,local outlier factor(LOF)and random forest(RF)fault detection and diagnosis model is proposed.The specific work is as follows:(1)The data running in the refrigeration system has the characteristics of non-linear,nonGaussian distribution and noise.This paper proposes a fault detection based on the LOF model.This method expresses the relationship between faults and normal with the meaning of data density,avoids the assumption of Gaussian distribution and data linearity,and obtains the control limits of LOF statistics through kernel density estimation(KDE).Taking ASHRAE RP-1043 chiller as the object,the 7 types of typical faults were tested by experimental simulation,and the simulation results were compared with PCA and OCSVM.The simulation results showed that the model detection rate using KDE-LOF was better than the other two models.When the fault level is high,the detection accuracy rate is greater than 90%.The model has a detection rate of greater than 83% and a false negative rate of less than 3%.(2)Due to the poor diagnostic performance of a single classifier,the ensemble learning model of random forest(RF)is introduced to improve the diagnostic effect of the refrigeration system.Combined with the particle swarm search algorithm(PSO),a fault diagnosis model based on PSORF is proposed.Taking ASHRAE RP-1043 chiller as the object,the 7 types of typical faults are simulated.The diagnosis results are compared with support vector machines(SVM),artificial neural networks(ANN)and a single DT model.The results show that the proposed PSO-RF diagnosis model has better diagnostic performance,and the diagnostic accuracy rate reaches 96.3% when the fault level is 1.(3)In order to verify the practical application effect of the fault diagnosis method,a fault simulation test bench was built by refitting a heat pump split type floor-standing air conditioner to simulate four common faults,compressor blow-by,condenser surface blockage,four-way Valve leakage and refrigerant leakage,analyze the change status of each parameter of the system when 4types of faults occur,the diagnosis results show that the use of LOF-RF model,the accuracy of detection of the four types of faults are greater than 88.6%,the overall diagnosis of the four types of faults is correct The rate is 99.7%,and the fault diagnosis of refrigeration and air-conditioning units is realized.