Train air conditioner refrigerant leakage fault prediction method based on XGBoost algorithm

Abstract

The invention discloses a train air conditioner refrigerant leakage fault prediction method based on an XGBoost algorithm, and belongs to the technical field of railway vehicle air conditioners, and the method comprises the following steps: A, data acquisition: carrying out a refrigerant leakage test, and collecting test data; step B, feature screening: performing correlation analysis and sorting on the data to obtain feature variables of an XGBoost model; step C, establishing an XGBoost model: establishing the XGBoost model through the screened characteristic variables; and step D, importing the data of the train air conditioning system controller into the XGBoost model, and carrying out refrigerant leakage fault prediction. According to the method, the detection effect on refrigerant leakage is good, refrigerant amount fault early warning is achieved through XGBoost, the state of the refrigerant amount of the air conditioning system is correspondingly indicated, maintenance is convenient, and the maintenance cost is reduced.

Description

Technical Field

[0001] This invention belongs to the field of rail vehicle air conditioning technology, specifically relating to a method for predicting refrigerant leakage faults in train air conditioning based on the XGBoost algorithm. Background Technology

[0002] The cooling principle of a train's air conditioning system: The refrigerant is compressed into high-temperature, high-pressure vapor by the compressor, enters the air-cooled condenser, and is condensed into a normal-temperature, high-pressure liquid by forced cooling from the outside air. After being throttled and depressurized by the thermostatic expansion valve, it is transformed into a low-temperature, low-pressure gas-liquid mixture, and then enters the evaporator. It absorbs heat from the air passing outside the evaporator and evaporates into low-temperature, low-pressure vapor, which is then drawn into the compressor, completing one refrigeration cycle. The compressor works continuously to achieve a continuous cooling effect.

[0003] Air conditioning unit refrigeration system malfunctions are generally not easily visualized, nor can the components be disassembled one by one. Only external inspections and a comprehensive analysis of the causes of the malfunction are possible. Traditional methods for diagnosing refrigerant leaks rely on analyzing the system's pressure; when the operating pressure exceeds the normal range, a malfunction is almost certain, indicating a significant refrigerant leak. Therefore, it is necessary to develop a method for predicting refrigerant leaks in train air conditioning systems. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm, which overcomes the shortcomings of the lack of existing refrigerant leakage detection methods. Through correlation analysis, it utilizes low-cost and collectable parameters to achieve fault early warning of refrigerant leakage.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for predicting refrigerant leakage faults in train air conditioning based on the XGBoost algorithm, comprising the following steps: Step A, Data Acquisition: Conduct a refrigerant leakage test on the air conditioning unit on the test bench and collect the test data; Step B, Feature Selection: Perform correlation analysis and sort the experimental data to obtain the relevant input variables for the XGBoost model; Step C: Build the XGBoost model: Build the XGBoost model using the relevant input variables obtained after filtering; Step D, Refrigerant Leakage Fault Prediction: Import the train air conditioning system controller data into the XGBoost model to predict refrigerant leakage faults.

[0006] Furthermore, in step A, data acquisition includes the following steps: A.1. Construct a test bench to simulate the experimental environment of refrigerant leakage failure; A.2 Before the test begins, calibrate the absolute time of the air conditioning unit controller (KPC) and the test bench to ensure that the two times are consistent and manually record the test conditions for each day; A.3. The refrigerant charge in the air conditioning unit starts from 110% charge and decreases in increments of 0.2 kg to 70% charge. Record the test operating data for each charge level. A.4 After the test, download the data from the air conditioning unit controller (KPC), the test data from the test bench, and save the manually recorded data.

[0007] Furthermore, in step A.3, when conducting tests at each charge level, A.31. First, fix the refrigerant charge and the fresh air valve opening and closing status. Under the outdoor dry bulb temperature of 23℃, 29℃, 35℃, 40℃ and 45℃ respectively, conduct the test according to each value of indoor dry bulb temperature of 21~26℃ and collect the test data. A.32. Change the status of the fresh air valve and repeat step A.31.

[0008] Furthermore, the air conditioning unit will shut down during the refrigerant reduction process and when the outdoor dry-bulb temperature, indoor dry-bulb temperature, and fresh air valve switching status are changed.

[0009] Furthermore, in step B, a parameter correlation analysis is performed on the experimental data. The correlation analysis includes... Calculate the correlation coefficient between the input variables and the refrigerant leakage rate, and then select the relevant input variables with a correlation coefficient greater than or equal to 0.9. In the formula, x is the input variable, and Y is the refrigerant dose. The input variables include the original data collected from the test bench, which include indoor inlet wet-bulb temperature, indoor inlet dry-bulb temperature, outdoor inlet dry-bulb temperature, outdoor inlet wet-bulb temperature, living room temperature, target temperature, indoor outlet wet-bulb temperature, intake temperature, exhaust temperature, high pressure, low pressure, blower current, condenser current, compressor current, compressor casing temperature, indoor air volume, evaporator outlet temperature, unit total voltage, unit total current, condenser outlet temperature, condenser outlet temperature, expansion valve inlet temperature, power frequency, indoor cooling capacity, indoor sensible heat, indoor latent heat, unit active power, unit power factor, unit apparent power, blower power factor, compressor power factor, and condenser power factor. The relevant input variables include high pressure, low pressure, exhaust superheat, indoor sensible heat, real-time superheat, intake temperature, exhaust temperature, expansion valve opening, compressor current, and fan current.

[0010] Furthermore, combining engineering experience and using correlation analysis again, the correlation between the relevant input variable parameters and the parameters that the air conditioning unit controller (KPC) can collect is calculated. The parameters that the air conditioning unit controller (KPC) cannot collect among the 10 relevant input variables are replaced with the parameters that can be collected from the air conditioning unit controller (KPC) parameters, resulting in 6 relevant variables: high pressure, low pressure, exhaust temperature, suction temperature, compressor current, and evaporator fan current. Based on engineering application experience, three engineering experience variables—subcooling, system pressure ratio, and temperature difference—were added to obtain nine relevant input variables.

[0011] Furthermore, step C, establishing the XGBoost model, includes the following steps: C.1. Divide the sample set of relevant input variables, with 80% used as training samples and 20% as test samples; C.2. Establish the objective function, input the training samples into the XGBoost model for training, and obtain the correspondence between the input variables and the refrigerant leakage amount, as well as the refrigerant leakage amount. C.3. Based on the ratio between the predicted refrigerant leakage and the standard charge of the train's air conditioning system, the corresponding refrigerant leakage range is obtained. C.4. Validate the model using test samples; C.5. Repeat steps C.1-C.4, and increase or decrease the relevant input variables to achieve a model accuracy of over 95%.

[0012] Furthermore, in the XGBoost model, the learning rate η, which is the Taylor series expansion of the loss function, ranges from 0.01 to 0.2.

[0013] Furthermore, the number of tree models K in the XGBoost model ranges from 100 to 200.

[0014] Furthermore, the maximum depth of each tree in the XGBoost model ranges from 3 to 10.

[0015] Furthermore, the refrigerant leakage range includes four intervals: [100%-90%], [89%-80%], [79%-70%], and [below 69%].

[0016] Compared to traditional methods of manual inspection or subsequent logical analysis of data, this invention demonstrates superior detection of refrigerant leaks. It utilizes XGBoost to provide early warning of refrigerant charge faults, indicating the status of the refrigerant charge in the air conditioning system, facilitating maintenance and reducing costs. Furthermore, a user-friendly web view framework has been developed, allowing direct import of air conditioning system data to analyze the refrigerant charge status. This approach offers advantages such as flexibility, direct applicability, time savings, and increased efficiency.

[0017] The present invention will now be described in detail with reference to the accompanying drawings. Attached Figure Description

[0018] Figure 1 This is a flowchart of a refrigerant leak fault prediction method based on the XGBoost algorithm; Figure 2 This is the interface of a web-based refrigerant leak fault early warning platform. Detailed Implementation

[0019] See appendix Figure 1 This invention provides a train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm, which is applicable to variable frequency air conditioners equipped with high and low pressure sensors, compressor exhaust temperature and suction temperature sensors, and capable of collecting compressor and blower current, or rail transit air conditioning units equipped with compressor and blower current sensors.

[0020] Specifically, the rail transit air conditioning unit includes components such as a variable frequency compressor, condenser, condenser fan, evaporator, evaporator fan, dryer filter, high pressure switch, low pressure switch, high pressure sensor, low pressure sensor, sight glass, exhaust temperature sensor, and suction temperature sensor. The refrigeration system is connected by welding copper pipes, and the refrigerant is R407C.

[0021] The method of the present invention includes the following steps.

[0022] Step A, Data Acquisition: Conduct a refrigerant leakage test on the air conditioning unit on the test bench and collect the test data.

[0023] Data acquisition specifically includes the following steps.

[0024] A.1. Construct a test bench to simulate the experimental environment of refrigerant leakage failure.

[0025] A.2 Before the test begins, calibrate the absolute time of the air conditioning unit controller KPC and the test bench to ensure that the two times are consistent and manually record the test conditions for each day.

[0026] A.3. The refrigerant charge in the air conditioning unit starts from 110% charge and decreases in increments of 0.2 kg to 70% charge. Record the test operating data for each charge level.

[0027] In step A.3, when conducting tests at each charge level, A.31. First, fix the refrigerant charge and the fresh air valve opening and closing status. Under the condition that the outdoor dry bulb temperature is 23℃, 29℃, 35℃, 40℃ and 45℃ respectively, conduct the test according to each value of indoor dry bulb temperature from 21 to 26℃ (i.e., 21℃, 22℃, 23℃, 24℃, 25℃ and 26℃, a total of 6 values), and collect the test data. A.32. Change the status of the fresh air valve and repeat step A.31.

[0028] A.4 After the test, download the data from the air conditioning unit controller KPC, the test data from the test bench, and save the manually recorded data.

[0029] The air conditioning unit will shut down when the refrigerant is reduced, or when the outdoor dry-bulb temperature, indoor dry-bulb temperature, and fresh air valve status are changed.

[0030] During the above tests, the air conditioning unit operated in cooling mode. Heating was achieved by using electric heating to heat the fresh air. In heating mode, only the blower and electric heating were working. The electric heating raised the air temperature, and the blower delivered the hot air into the vehicle. It was not a heat pump unit heating method. Therefore, the refrigerant leakage test was only simulated in the air conditioning unit's cooling mode.

[0031] During the test, the refrigerant leak location was determined according to common leak locations (pressure valve). The air conditioning unit compressor frequency was 40-70Hz during the test, and a fixed frequency was used for the test. In actual vehicle operation, the cooling effect is affected by the fresh air volume, which is affected by speed. During the test, the effect of speed on the fresh air volume was simulated by controlling the fresh air valve to be fully open or fully closed. The test involved four control variables: refrigerant charge, fresh air valve opening / closing, outdoor dry-bulb temperature, and indoor dry-bulb temperature, totaling 300 test conditions. The outdoor dry-bulb temperature simulated the fresh air temperature, and the indoor dry-bulb temperature simulated the mixed temperature (return air and fresh air). The humidity was set to 60%RH.

[0032] Step B, Feature Selection: Perform correlation analysis and sort the experimental data to obtain the relevant input variables for the XGBoost model.

[0033] In step B, a parametric correlation analysis is performed on the experimental data. The correlation analysis includes... Calculate the correlation coefficient between the input variables and the refrigerant leakage rate, and then select the relevant input variables with a correlation coefficient greater than or equal to 0.9. In the formula, x is the input variable, and Y is the refrigerant dose. Input variables include the raw data collected from the test bench, which include: indoor inlet wet-bulb temperature, indoor inlet dry-bulb temperature, outdoor inlet dry-bulb temperature, outdoor inlet wet-bulb temperature, passenger room temperature, target temperature, indoor outlet wet-bulb temperature, intake temperature, exhaust temperature, high pressure, low pressure, blower current, condenser current, compressor current, compressor casing temperature, indoor air volume, evaporator outlet temperature, unit total voltage, unit total current, condenser outlet temperature, condenser outlet temperature, expansion valve inlet temperature, power frequency, indoor cooling capacity, indoor sensible heat, indoor latent heat, unit active power, unit power factor, unit apparent power, blower power factor, compressor power factor, and condenser power factor.

[0034] The relevant input variables with a correlation coefficient greater than or equal to 0.9 are 10 variables in total: high pressure, low pressure, exhaust superheat, indoor sensible heat, real-time superheat, intake temperature, exhaust temperature, expansion valve opening, compressor current, and fan current.

[0035] The correlation analysis method was used to calculate the correlation between the parameters of each variable and the parameters that the KPC controller of the air conditioning unit can collect. The parameters that the KPC cannot collect in the 10 related variables were replaced with the parameters that the KPC can collect, and finally, 6 related variables were obtained: high pressure, low pressure, exhaust temperature, suction temperature, compressor current and evaporator fan current. Based on engineering experience, three engineering experience variables—subcooling, system pressure ratio, and temperature difference—were added to obtain nine relevant input variables.

[0036] Step C: Build the XGBoost model: Build the XGBoost model using the relevant input variables obtained after filtering.

[0037] Building an XGBoost model involves the following steps: C.1. Divide the sample set of relevant input variables, with 80% used as training samples and 20% as test samples; C.2. Establish the objective function, input the training samples into the XGBoost model for training, and obtain the correspondence between the input variables and the refrigerant leakage amount, as well as the refrigerant leakage amount. C.3. Based on the ratio between the predicted refrigerant leakage and the standard charge of the train's air conditioning system, the corresponding refrigerant leakage range is obtained. C.4. Validate the model using test samples; C.5. Repeat steps C.1-C.4, and increase or decrease the relevant input variables to make the model accuracy reach more than 95%. Finally, the nine input variables are determined to be: high pressure, low pressure, exhaust temperature, intake temperature, compressor current, evaporator fan current, subcooling, system pressure ratio and temperature difference.

[0038] The XGBoost algorithm can learn to handle missing values ​​and effectively prevent overfitting by adding a regularization term. Therefore, it has a strong ability to adapt to new data and can adapt well to test data of train air conditioning system controller and mainline operation data.

[0039] Based on the mechanism analysis of refrigerant leakage in air conditioning systems, the refrigerant leakage process is a slow-changing process, but it involves many input signals. Therefore, a regression equation between multiple input signals and leakage amount was trained using XGBoost. At the same time, in order to ensure the reliability of the leakage amount warning, the predicted leakage amount was divided into different intervals of refrigerant amount.

[0040] XGBoost's predictions are the sum of the predictions of each tree in its model, that is, the sum of the scores of the corresponding leaf nodes of each tree: .

[0041] The objective function of the XGBoost model: Based on the objective function, learn the following... K A tree model That is, training the model: .

[0042] in, This represents the model's predicted value. Indicates the first The category label of each sample, K Indicates the number of trees, Let represent the model of the k-th tree, T represent the number of leaf nodes in each tree, and w represent the set of scores for the leaf nodes in each tree. and Represents the coefficient.

[0043] Simplified objective function: .

[0044] Splitting index of the gain formula: .

[0045] Based on the above XGBoost algorithm principle, refrigerant leakage fault prediction is as follows: Figure 1As shown, a correlation analysis is first performed on the original refrigerant leakage test data. Based on the correlation analysis results, associated variables are selected for model training. Based on the model performance, it is considered whether to add / remove associated variables in the model in combination with engineering experience and correlation analysis results. This process is repeated iteratively until the model achieves an accuracy of over 95%.

[0046] The XGBoost algorithm is used to predict refrigerant dosage. An objective function for the XGBoost model is established, comprising a loss function and a structure function. Trees are continuously added to the model, with feature splitting used to grow each tree. The model is trained to obtain the correspondence between input variables and refrigerant leakage, and the refrigerant leakage amount is calculated. Based on the ratio between the predicted refrigerant leakage amount and the standard charge amount of the train's air conditioning system, the corresponding refrigerant leakage range is obtained. Specifically, the Taylor series expansion learning rate η in the objective function ranges from 0.01 to 0.2. The number of tree models K ranges from 100 to 200. The maximum depth of each tree ranges from 3 to 10.

[0047] Step D, Refrigerant Leakage Fault Prediction: Import the train air conditioning system controller data into the XGBoost model to predict refrigerant leakage faults.

[0048] Step E: To achieve visualization, a refrigerant leak fault early warning platform needs to be developed. This involves building a web application framework based on Django, which allows for the import of offline data from the train air conditioning system controller and the early warning of refrigerant leak faults through the web interface.

[0049] When the platform is running, such as Figure 2 As shown, click "Select File" to import controller data. After clicking the "Leak Detection" button, the refrigerant leak fault early warning algorithm model will run to detect the imported controller data, detecting the refrigerant dosage range (mainly including four ranges: [100%-90%], [89%-80%], [79%-70%], and [below 69%]). The refrigerant dosage of the imported data will be displayed in the results display box on the right. The page will display the system's refrigerant dosage range status according to the start time of the same refrigerant dosage range. If the imported data contains multiple refrigerant dosage ranges, the results will be displayed in multiple lines according to the above format. The detection results can be downloaded and saved as needed.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A method for predicting refrigerant leakage faults in train air conditioning systems based on the XGBoost algorithm, characterized in that, Includes the following steps: Step A, Data Acquisition: Conduct a refrigerant leakage test on the air conditioning unit on the test bench and collect the test data; Step B, Feature Selection: Perform correlation analysis and sort the experimental data to obtain the relevant input variables for the XGBoost model; Step C: Build the XGBoost model: Build the XGBoost model using the relevant input variables obtained after filtering; Step D, Refrigerant Leakage Fault Prediction: Import the train air conditioning system controller data into the XGBoost model to predict refrigerant leakage faults.

2. The train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm according to claim 1, characterized in that, In step A, data acquisition includes the following steps: A.

1. Construct a test bench to simulate the experimental environment of refrigerant leakage failure; A.2 Before the test begins, calibrate the absolute time of the air conditioning unit controller (KPC) and the test bench to ensure that the two times are consistent and manually record the test conditions for each day; A.

3. The refrigerant charge in the air conditioning unit starts from 110% charge and decreases in increments of 0.2 kg to 70% charge. Record the test operating data for each charge level. A.4 After the test, download the data from the air conditioning unit controller (KPC), the test data from the test bench, and save the manually recorded data.

3. The train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm according to claim 4, characterized in that, In step A.3, when conducting tests at each charge level, A.

31. First, fix the refrigerant charge and the fresh air valve opening and closing status. Under the outdoor dry bulb temperature of 23℃, 29℃, 35℃, 40℃ and 45℃ respectively, conduct the test according to each value of indoor dry bulb temperature of 21~26℃ and collect the test data. A.

32. Change the status of the fresh air valve and repeat step A.

31.

4. The train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm according to claim 4, characterized in that, The air conditioning unit will shut down when the refrigerant is reduced, or when the outdoor dry-bulb temperature, indoor dry-bulb temperature, and fresh air valve status are changed.

5. The train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm according to any one of claims 1-4, characterized in that, In step B, a parametric correlation analysis is performed on the experimental data. The correlation analysis includes... Calculate the correlation coefficient between the input variables and the refrigerant leakage rate, and then select the relevant input variables with a correlation coefficient greater than or equal to 0.

9. In the formula, x is the input variable, and Y is the refrigerant dose. The input variables include the original data collected from the test bench, which include indoor inlet wet-bulb temperature, indoor inlet dry-bulb temperature, outdoor inlet dry-bulb temperature, outdoor inlet wet-bulb temperature, living room temperature, target temperature, indoor outlet wet-bulb temperature, intake temperature, exhaust temperature, high pressure, low pressure, blower current, condenser current, compressor current, compressor casing temperature, indoor air volume, evaporator outlet temperature, unit total voltage, unit total current, condenser outlet temperature, condenser outlet temperature, expansion valve inlet temperature, power frequency, indoor cooling capacity, indoor sensible heat, indoor latent heat, unit active power, unit power factor, unit apparent power, blower power factor, compressor power factor, and condenser power factor. The relevant input variables include high pressure, low pressure, exhaust superheat, indoor sensible heat, real-time superheat, intake temperature, exhaust temperature, expansion valve opening, compressor current, and fan current.

6. The train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm according to claim 5, characterized in that, Based on engineering experience and by using correlation analysis again, the correlation between relevant input variable parameters and parameters that can be collected by the air conditioning unit controller (KPC) is calculated. The parameters that cannot be collected by the air conditioning unit controller (KPC) among the 10 relevant input variables are replaced by the parameters that can be collected by the air conditioning unit controller (KPC), resulting in 6 relevant variables: high pressure, low pressure, exhaust temperature, suction temperature, compressor current, and evaporator fan current. Based on engineering application experience, three engineering experience variables—subcooling, system pressure ratio, and temperature difference—were added to obtain nine relevant input variables.

7. The train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm according to claim 6, characterized in that, Step C, which involves building the XGBoost model, includes the following steps: C.

1. Divide the sample set of relevant input variables, with 80% used as training samples and 20% as test samples; C.

2. Establish the objective function, input the training samples into the XGBoost model for training, and obtain the correspondence between the input variables and the refrigerant leakage amount, as well as the refrigerant leakage amount. C.

3. Based on the ratio between the predicted refrigerant leakage and the standard charge of the train's air conditioning system, the corresponding refrigerant leakage range is obtained. C.

4. Validate the model using test samples; C.

5. Repeat steps C.1-C.4, and increase or decrease the relevant input variables to achieve a model accuracy of over 95%.

8. The train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm according to claim 7, characterized in that, In the XGBoost model, the learning rate η, which is the Taylor series expansion of the loss function, ranges from 0.01 to 0.

2.

9. The train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm according to claim 7, characterized in that, In the XGBoost model, the number of tree models K ranges from 100 to 200.

10. The train air conditioning refrigerant leakage fault prediction method based on the XGBoost algorithm according to claim 7, characterized in that, The refrigerant leakage range includes four intervals: [100%-90%], [89%-80%], [79%-70%], and [below 69%].

Images