Search

CN-121994412-A - Train air conditioner refrigerant leakage fault prediction method based on XGBoost algorithm

CN121994412ACN 121994412 ACN121994412 ACN 121994412ACN-121994412-A

Abstract

The invention discloses a refrigerant leakage fault prediction method for a train air conditioner based on XGBoost algorithm, which belongs to the technical field of railway vehicle air conditioners and comprises the following steps of A, obtaining data, collecting test data, B, performing feature screening, namely performing correlation analysis on the data and sequencing to obtain feature variables of a XGBoost model, C, building a XGBoost model, namely building a XGBoost model through the screened feature variables, and D, importing data of a train air conditioner system controller into the XGBoost model to predict refrigerant leakage faults. The method has good detection effect on refrigerant leakage, and can realize refrigerant quantity fault early warning by applying XGBoost, give corresponding indication for the state of the refrigerant quantity of the air conditioning system, facilitate maintenance and reduce maintenance cost.

Inventors

  • WU HANSHENG
  • LIU BIN
  • Luo Shuihu
  • FENG XUWEI
  • ZHANG YANAN

Assignees

  • 石家庄国祥运输设备有限公司

Dates

Publication Date
20260508
Application Date
20260105

Claims (10)

  1. 1. A train air conditioner refrigerant leakage fault prediction method based on XGBoost algorithm is characterized by comprising the following steps: Step A, data acquisition, namely performing a refrigerant leakage test on an air conditioning unit on a test bed, and collecting test data; step B, feature screening, namely performing correlation analysis and sequencing on test data to obtain relevant input variables of XGBoost models; Step C, establishing XGBoost a model, namely establishing XGBoost model through the related input variables obtained after screening; And D, refrigerant leakage fault prediction, namely importing data of a train air conditioning system controller into a XGBoost model to perform refrigerant leakage fault prediction.
  2. 2. The method for predicting refrigerant leakage failure of a train air conditioner based on XGBoost algorithm as recited in claim 1, wherein in step a, the data acquisition includes the steps of: a.1, building a test bed for simulating a refrigerant leakage fault experimental environment; a2, calibrating absolute time of an air conditioning unit controller (KPC) and a test bed before starting a test, enabling the absolute time of the KPC and the absolute time of the test bed to be consistent, and manually recording test working conditions of each day; The refrigerant filling amount in the air conditioning unit is gradually decreased from 110% to 70% according to a gradient of 0.2kg, and test working condition data under each filling amount are recorded; and A.4, downloading data of an air conditioning unit controller (KPC), test data of a test bed and manually recorded data after the test is finished.
  3. 3. The method for predicting refrigerant leakage failure in a train air conditioner based on XGBoost algorithm according to claim 4, wherein, when the test is performed at each charge amount in step a.3, A.31, firstly fixing the refrigerant filling amount and the fresh air valve switching state, and then carrying out a test according to each value of the indoor dry bulb temperature of 21-26 ℃ under the conditions that the outdoor dry bulb temperature is 23 ℃, 29 ℃,35 ℃, 40 ℃ and 45 ℃ respectively, and collecting test data; And A.32, changing the on-off state of the fresh air valve, and repeating the step A.31.
  4. 4. The method for predicting refrigerant leakage failure of a train air conditioner based on XGBoost algorithm as recited in claim 4, wherein the air conditioning unit is stopped during a refrigerant reduction process and when an outdoor dry bulb temperature, an indoor dry bulb temperature and a fresh air valve on-off state are changed.
  5. 5. The method for predicting refrigerant leakage failure of air conditioner in train based on XGBoost algorithm as recited in any one of claims 1 to 4, wherein the step B of performing a parameter correlation analysis on the test data includes the step of performing a correlation analysis based on the result of the parameter correlation analysis Calculating a correlation coefficient between the input variable and the refrigerant leakage amount, and then screening out a correlation input variable having a correlation coefficient of 0.9 or more, In the formula, x is an input variable, Y is the refrigerant quantity, The input variables comprise original acquisition variables of a test bed, wherein the original acquisition variables comprise indoor air inlet wet bulb temperature, indoor air inlet dry bulb temperature, outdoor air inlet wet bulb temperature, guest room temperature, target temperature, indoor air outlet wet bulb temperature, air suction temperature, exhaust temperature, high-pressure, low-pressure, blower current, condensing fan current, compressor shell temperature, indoor air quantity, evaporator outlet temperature, unit total voltage, unit total current, condensing outlet temperature, condenser air outlet temperature, expansion valve front temperature, power frequency, indoor refrigerating capacity, indoor heat development, indoor latent heat, unit active power, unit power factor, unit apparent power, blower power factor, compressor power factor and condensing fan power factor; The relevant input variables include high pressure, low pressure, exhaust superheat, indoor heat development, real-time superheat, suction temperature, exhaust temperature, expansion valve opening, compressor current and ventilator current.
  6. 6. The method for predicting leakage faults of air conditioning refrigerant in trains based on XGBoost algorithm according to claim 5, wherein the correlation degree between the parameters of the relevant input variables and the parameters which can be collected by the air conditioning unit controller (KPC) is calculated by combining engineering experience and reusing correlation analysis, and the parameters which can not be collected by the air conditioning unit controller (KPC) in 10 relevant input variables are replaced by the parameters which can be collected by the air conditioning unit controller (KPC) to obtain 6 relevant variables in total, namely high pressure, low pressure, exhaust temperature, suction temperature, compressor current and evaporating fan current; and combining engineering application experience, and adding 3 engineering experience variables of supercooling degree, system pressure ratio and temperature difference to obtain 9 relevant input variables.
  7. 7. The method for predicting refrigerant leakage failure of a train air conditioner based on XGBoost algorithm as recited in claim 6, wherein the step C of creating XGBoost model includes the steps of: c.1, dividing a sample set of related input variables, wherein 80% of the sample set is used as training samples, and 20% of the sample set is used as test samples; c.2, establishing an objective function, inputting a training sample into a XGBoost model for training, and obtaining a corresponding relation between an input variable and the refrigerant leakage; C.3, obtaining a corresponding refrigerant leakage interval according to the duty ratio between the predicted refrigerant leakage and the standard filling amount of the train air conditioning system; C.4, performing model verification by using a test sample; And C.5, repeating the steps C.1-C.4, and increasing or decreasing related input variables to enable the model accuracy to reach more than 95%.
  8. 8. The method for predicting refrigerant leakage failure of a train air conditioner based on XGBoost algorithm as set forth in claim 7, wherein the loss function taylor series expansion learning rate η in the XGBoost model is in a range of 0.01 to 0.2.
  9. 9. The method for predicting refrigerant leakage failure of train air conditioner based on XGBoost algorithm as set forth in claim 7, wherein the number K of tree models in the XGBoost model is 100-200.
  10. 10. The method for predicting refrigerant leakage failure of a train air conditioner based on XGBoost algorithm according to claim 7, wherein the refrigerant leakage amount section includes four sections of [100% -90% ], [89% -80% ], [79% -70% ], and [69% or less ].

Description

Train air conditioner refrigerant leakage fault prediction method based on XGBoost algorithm Technical Field The invention belongs to the technical field of rail vehicle air conditioners, and particularly relates to a train air conditioner refrigerant leakage fault prediction method based on XGBoost algorithm. Background The refrigerating principle of the train air conditioning system is that the refrigerant is compressed into high-temperature and high-pressure steam by a compressor, enters an air-cooled condenser, is subjected to forced cooling by external air, is condensed into normal-temperature and high-pressure liquid, is throttled and depressurized by a thermal expansion valve, is converted into a low-temperature and low-pressure gas-liquid mixed state, then enters an evaporator, absorbs the heat of the air passing through the outer side of the evaporator, is evaporated into low-temperature and low-pressure steam, is sucked by the compressor, and completes a refrigerating cycle, and the compressor continuously works to achieve the effect of continuous refrigeration. The fault of the refrigerating system of the air conditioning unit generally cannot directly see the position where the fault occurs, cannot separate the components of the refrigerating system one by one, can only perform appearance inspection, and can comprehensively analyze the cause of fault production. The traditional method for judging the refrigerant leakage is to analyze the pressure of the refrigeration system, and the refrigerant leakage has taken a serious step when the operation pressure is out of the normal range and the fault exists. Therefore, it is necessary to develop a train air conditioner refrigerant leakage failure prediction method. Disclosure of Invention The technical problem to be solved by the invention is to provide a train air conditioner refrigerant leakage fault prediction method based on XGBoost algorithm, which overcomes the defect of lack of the existing refrigerant leakage detection means, and realizes the fault early warning of refrigerant leakage by utilizing low-cost acquirable parameters through correlation analysis. In order to solve the technical problems, the technical scheme adopted by the invention is that the train air conditioner refrigerant leakage fault prediction method based on XGBoost algorithm comprises the following steps: Step A, data acquisition, namely performing a refrigerant leakage test on an air conditioning unit on a test bed, and collecting test data; step B, feature screening, namely performing correlation analysis and sequencing on test data to obtain relevant input variables of XGBoost models; Step C, establishing XGBoost a model, namely establishing XGBoost model through the related input variables obtained after screening; And D, refrigerant leakage fault prediction, namely importing data of a train air conditioning system controller into a XGBoost model to perform refrigerant leakage fault prediction. Further, in step a, the data acquisition includes the steps of: a.1, building a test bed for simulating a refrigerant leakage fault experimental environment; a2, calibrating absolute time of an air conditioning unit controller (KPC) and a test bed before starting a test, enabling the absolute time of the KPC and the absolute time of the test bed to be consistent, and manually recording test working conditions of each day; The refrigerant filling amount in the air conditioning unit is gradually decreased from 110% to 70% according to a gradient of 0.2kg, and test working condition data under each filling amount are recorded; and A.4, downloading data of an air conditioning unit controller (KPC), test data of a test bed and manually recorded data after the test is finished. Further, when the test is performed at each of the charges in step A.3, A.31, firstly fixing the refrigerant filling amount and the fresh air valve switching state, and then carrying out a test according to each value of the indoor dry bulb temperature of 21-26 ℃ under the conditions that the outdoor dry bulb temperature is 23 ℃, 29 ℃,35 ℃, 40 ℃ and 45 ℃ respectively, and collecting test data; And A.32, changing the on-off state of the fresh air valve, and repeating the step A.31. Further, the air conditioning unit is shut down during the refrigerant reduction process and when the outdoor dry bulb temperature, the indoor dry bulb temperature and the fresh air valve on-off state are changed. Further, in step B, the test data is subjected to a parameter correlation analysis, the correlation analysis including a correlation analysis based on the dataCalculating a correlation coefficient between the input variable and the refrigerant leakage amount, and then screening out a correlation input variable having a correlation coefficient of 0.9 or more, In the formula, x is an input variable, Y is the refrigerant quantity, The input variables comprise original acquisition variables of a test bed, wherein the original a