Sensitivity Analysis And Prediction Of Refrigerant Leakage In The Air Conditioning System

Refrigerant leakage is one of the most common faults in the operation of the air conditioning system.This kind of fault will cause long-term abnormal operation of the system,waste of energy,environmental pollution,and the user’s thermal comfort needs can not be met,so it is of great significance to detect and diagnose it in time.In this thesis,a data mining refrigerant charge prediction model based on the sensitivity analysis method is proposed for the refrigerant leakage fault of the air conditioning system.According to the historical operation data of a 12 kW split air conditioning system,the prediction model and early warning mechanism are established.The results show that the prediction model and early warning mechanism based on sensitivity analysis method perform well,it can accurately predict the refrigerant charge,and timely warn the charge range in the leakage process.In this regard,the following work is carried out in this thesis:First of all,the refrigerant leakage experiments are divided into two groups,which are carried out in the high-precision enthalpy difference laboratory.One group is 34 groups of operation experiments of air conditioning system under two working conditions and 17 charging levels,and the other group is two groups of continuous leakage experiments under two working conditions,a total of 36 groups.During the experiment,the historical data were collected and monitored.After the experiment,the uncertainty of the experimental data was analyzed to verify the accuracy of the experimental data.Secondly,using a data mining algorithm to build a robust and reliable model,the key lies in whether the selection of feature variables is correct,so the method of selecting feature variables needs to be accurate and effective.In this thesis,the sensitivity analysis of the air conditioning system at the system and parameter level is proposed.Combined with professional knowledge,the characteristic variables sensitive to refrigerant leakage fault are selected as the input variables of the prediction model,which not only ensures the quality of the characteristic variables and makes the prediction model have high robustness,but also summarizes the optimal charging range of the system and makes the prediction model have professional interpretability.Finally,according to the characteristic variables selected by the sensitivity analysis,the Long Short-Term Memory neural network prediction model based on the improved virtual cooling capacity sensor is established.The results show that the model error of the improved virtual cooling capacity sensor is less than 5%,and the prediction model performs well and stably.Combined with the mutation detection algorithm and the optimal charging range,the charging warning mechanism is established,which can accurately predict the three levels of cooling capacity early warning of refrigerant leakage.The DALEX technology is used to explain and optimize the prediction model.The model explanation can organically integrate the expert knowledge and algorithm model,and enhance the robustness of the prediction model.