Method for predicting water temperature of outlet of locomotive converter cabinet body based on CNN (Convolutional Neural Network) model

Abstract

The invention relates to the field of locomotive electric power systems, in particular to a locomotive converter cabinet outlet water temperature prediction method based on a CNN model, and the method specifically comprises the following steps: collecting historical data related to the converter cabinet outlet water temperature, preprocessing the collected data to form a data set, and dividing the data set into a training set, a verification set and a test set; constructing a CNN model; training and verifying the CNN model by using the training set and the verification set; evaluating the trained CNN model by using the test set; and the trained CNN model obtains real-time data related to the water temperature of the outlet of the converter cabinet body, and a fault is predicted and diagnosed by using an abnormality diagnosis algorithm based on the real-time data in the CNN model. The invention provides a deep learning-based method for predicting the water temperature of the outlet of the converter cabinet body of the locomotive, the water temperature of the outlet of the converter cabinet body is predicted by learning the characteristics of the input data, and an efficient, accurate and automatic temperature monitoring and predicting scheme is provided for a locomotive power system in this way.

Description

Technical Field

[0001] The present invention relates to the field of locomotive power systems, and in particular to a method for predicting the outlet water temperature of a locomotive converter cabinet based on a CNN model. Background Art

[0002] In a locomotive environment, the stable operation of the power system is crucial to the safety and performance of the train. As one of the core devices for power conversion, the operating status of the converter cabinet directly affects the power supply and performance of the entire train. Therefore, accurate prediction and monitoring of the outlet water temperature of the converter cabinet is of great significance in a locomotive environment.

[0003] The locomotive's power system is responsible for providing the train with power and the electrical energy required for operation. The converter cabinet plays a key role in converting direct current into alternating current. Its normal operation is directly related to the train's traction performance and the reliability of the system.

[0004] Power equipment generates heat during operation, so the outlet water temperature of the converter cabinet is an important operating parameter. Excessive temperature may cause equipment overheating, performance degradation, or even damage to the equipment. Therefore, accurate prediction and monitoring of the temperature of the converter cabinet is a necessary task to ensure the safe operation of the power system. In the locomotive environment, real-time performance is crucial. Early detection of temperature anomalies and timely measures can avoid long-term overload or overheating of equipment and ensure the stability of the power system and the safety of the train. Summary of the invention

[0005] The present invention aims to solve the above-mentioned defects and provides a method for predicting the outlet water temperature of a locomotive converter cabinet based on a CNN model.

[0006] In order to overcome the defects in the background technology, the technical solution adopted by the present invention to solve the technical problem is: a method for predicting the outlet water temperature of a locomotive converter cabinet based on a CNN model, the specific steps are as follows:

[0007] S1: Collect historical data related to the outlet water temperature of the converter cabinet, pre-process the collected data to form a data set, and divide the data set into a training set, a validation set, and a test set;

[0008] S2: Build CNN model;

[0009] S3: Use the training set and validation set to train and validate the CNN model;

[0010] S4: Use the test set to evaluate the trained CNN model;

[0011] S5: The trained CNN model obtains real-time data related to the outlet water temperature of the converter cabinet, and uses the abnormal diagnosis algorithm based on real-time data in the CNN model to predict and diagnose faults.

[0012] A further improvement includes that in step S1, preprocessing includes data cleaning, normalization processing and outlier processing.

[0013] A further improvement includes that in step S2, the CNN model includes an input layer 1, convolutional layers 1, 2, a pooling layer 1, a convolutional layer 3, a pooling layer 2, a flattening layer 1, and connection layers 1, 2.

[0014] A further improvement includes that in step S5, the abnormality diagnosis algorithm calculation includes the following steps:

[0015] 4.1. If the predicted outlet water temperature of the converter cabinet is greater than the alarm temperature threshold, the CNN model determines that it is abnormal and an alarm is issued. Otherwise, the predicted outlet water temperature of the converter cabinet is compared with the actual outlet water temperature of the converter cabinet.

[0016] 4.2. If the difference between the predicted converter cabinet outlet water temperature and the actual converter cabinet outlet water temperature is greater than the threshold value, it is considered abnormal and an alarm is issued.

[0017] Further improvements include that the calculation of the threshold value includes the following steps:

[0018] 5.1. Based on the test set, the CNN model is used to predict the data set, and the difference between the predicted data set and the test set is calculated as diff;

[0019] 5.2. Calculate the vector norm of diff and sort it;

[0020] 5.3. Use the preset cutoff value of the sorted vector norm as the threshold.

[0021] A further improvement, included in step S3, comprises the following steps:

[0022] 6.1. Initialize the neural network model, specify filters and parameters / weights;

[0023] 6.2. The data in the training set is used as input for training, and the output layer data can be obtained, namely, the outlet water temperature of 1 converter cabinet and the outlet water temperature of 2 converter cabinets;

[0024] 6.3. Calculate the total error of the output layer, yi: actual value, Predicted value

[0025] Total error calculation

[0026]

[0027] 6.4 When the error does not meet the design goal, optimize until it meets the requirements, use the error back propagation algorithm and gradient descent method to solve and update the weights of the convolutional neural network model, update all filter values / weights and parameter values, iterate multiple times, and reduce the output error;

[0028] 6.5 Repeat steps 2-4 to complete the training of the neural network model.

[0029] The beneficial effects of the present invention are as follows: the present invention proposes a locomotive converter cabinet outlet water temperature prediction method based on a convolutional neural network (CNN) model. Aiming at the prediction demand of the locomotive converter cabinet outlet water temperature, a method based on deep learning is proposed. By learning the characteristics of the input data, especially the characteristics of the time series, the outlet water temperature of the converter cabinet is predicted. This converter cabinet outlet water temperature prediction method based on the CNN model combines the advantages of deep learning and provides a locomotive power system with an efficient, accurate and automated temperature monitoring and prediction solution;

[0030] High accuracy: The CNN model performs well in processing data with spatial structure characteristics, and can more accurately capture the complex variation of the outlet water temperature of the converter cabinet, thus improving the accuracy of the prediction.

[0031] Strong adaptability: The CNN model's ability to process time series data makes it more adaptable to the prediction of the outlet water temperature of the converter cabinet in the locomotive environment, and it can learn more representative features from the data;

[0032] Real-time: The deep learning model can learn and make rapid inferences in large-scale data, so it can achieve real-time prediction of the outlet water temperature of the converter cabinet, meeting the real-time requirements of the locomotive power system;

[0033] Automation: Through the trained CNN model, the outlet water temperature of the converter cabinet can be automatically monitored and predicted, which reduces the dependence on manual intervention and improves the automation level of the system;

[0034] Predictive maintenance: By predicting the trend of the outlet water temperature of the converter cabinet in advance, predictive maintenance of the power system can be achieved, potential problems can be discovered in time and corresponding measures can be taken, reducing the risk of sudden failures;

[0035] Safety assurance: Accurate monitoring of the water temperature at the converter cabinet outlet helps improve the safety of the power system. When temperature abnormalities occur, the system can take timely measures to ensure the safe operation of the train. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0037] Figure 1 It is the overall process of the present invention;

[0038] Figure 2 It is a flowchart of constructing a CNN model in the present invention;

[0039] Figure 3 It is the abnormality diagnosis logic diagram of the present invention;

[0040] Figure 4 This is the threshold value calculation method in the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] like Figure 1 As shown, a method for predicting the outlet water temperature of a locomotive converter cabinet based on a CNN model is shown, and the specific steps are as follows:

[0043] S1: Collect historical data related to the outlet water temperature of the converter cabinet, including the inlet water pressure of the converter cabinet, the outlet water pressure of the converter cabinet, the inlet water temperature of the converter cabinet, the outlet water temperature of the converter cabinet, and the air temperature of the converter cabinet. Preprocess the collected data to form a data set, and divide the data set into a training set, a validation set, and a test set, where 60% of the data set is used as a training set, 20% as a validation set, and 20% as a test set.

[0044] S2: Build CNN model;

[0045] S3: Use the training set and validation set to train and validate the CNN model. Use the training set to train the constructed CNN model and adjust the model parameters through the back propagation algorithm to make it better fit the data.

[0046] S4: Use the test set to evaluate the trained CNN model. This step is used to check the performance of the model on the validation set data. The evaluation indicator RMSE is used to measure the prediction accuracy of the model to ensure that the model has good generalization ability.

[0047] S5: The trained CNN model obtains real-time data related to the outlet water temperature of the converter cabinet, and uses the abnormal diagnosis algorithm based on real-time data in the CNN model to predict and diagnose faults.

[0048] In step S1, preprocessing includes data cleaning, normalization and outlier processing. Data cleaning includes data deduplication, missing value processing and erroneous data correction. Data deduplication is to identify completely duplicated data rows through the function of database query statements or data processing tools, and then judge based on specific key columns to remove duplicates and retain unique data records. Missing value processing is to check which fields in the data set have missing values, which can be determined by counting the number of null values ​​in each field. Error data correction is to check whether the data type or data value is wrong, transform the type or identify and correct the data value. Normalization is to scale the data to a specific interval, such as the [0,1] interval, so that the data is on the same scale, which is convenient for subsequent analysis and processing. Outlier processing is to keep a backup of the original data when cleaning the data, so as to backtrack and verify when necessary. At the same time, record the process and method of data cleaning so that others can understand and repeat the process.

[0049] In step S2, the CNN model includes input layer 1, convolutional layers 1, 2, pooling layer 1, convolutional layer 3, pooling layer 2, flattening layer 1 and connection layers 1, 2. The input and output of the CNN model are determined to ensure that the input contains sufficient information to accurately predict the outlet water temperature. The model parameter setting includes setting the number of network layers, convolution kernel size, pooling layer and other parameters, so as to make reasonable model parameter settings.

[0050] like Figure 2 As shown, the CNN (convolutional neural network) model includes:

[0051] 3.1 Input layer: The preprocessed sensor data is imported into the input layer. The input data is a vector sequence consisting of 300 14-dimensional vectors.

[0052] 3.2 Convolutional layer: The length of the convolution window used in convolutional layers 9, 10, and 11 is 3, the convolution step size is 1, and the data boundary is not filled;

[0053] 3.3 Pooling layer: Pooling layers 6 and 7 select maximum pooling, the pooling window size is 2, that is, the maximum value of the values ​​​​between 2 points is taken, and the sliding step size of the pooling window is 2;

[0054] 3.4 Activation function, the output after convolution is processed by ReLU activation function as the input of the next convolution layer. Compared with sigmoid, relu can effectively prevent the gradient diffusion problem and increase nonlinearity well;

[0055] 3.5 Flatten layer: Flatten layer 3 is used to "flatten" the input, that is, to convert multi-dimensional input into one dimension. It is often used in the transition from convolutional layer to fully connected layer.

[0056] 3.6 fully connected layer, the number of neurons in the fully connected layer 6 is 100, and the number of neurons in the fully connected layer 7 is 2 (i.e., the outlet water temperature of 1 converter cabinet and the outlet water temperature of 2 converter cabinets).

[0057] like Figure 3 As shown, in step S5, the abnormality diagnosis algorithm calculation includes the following steps:

[0058] 4.1. If the predicted outlet water temperature of the converter cabinet is greater than the alarm temperature threshold, the CNN model determines that it is abnormal and an alarm is issued. Otherwise, the predicted outlet water temperature of the converter cabinet is compared with the actual outlet water temperature of the converter cabinet.

[0059] 4.2. If the difference between the predicted converter cabinet outlet water temperature and the actual converter cabinet outlet water temperature is greater than the threshold value, it is considered abnormal and an alarm is issued.

[0060] according to Figure 4 As shown, the calculation of the threshold value includes the following steps:

[0061] 5.1. Based on the data set and test set predicted by the CNN model, calculate the difference between the predicted data set and the test set as diff;

[0062] 5.2. Calculate the vector norm of diff and sort it;

[0063] 5.3. Use the preset cutoff value of the sorted vector norm as the threshold.

[0064] In step S2, the following steps are included:

[0065] 6.1 Initialize the neural network model, specify filters and parameters / weights;

[0066] 6.2 The data in the training set is used as input for training, and the output layer data can be obtained, namely, the outlet water temperature of 1 converter cabinet and the outlet water temperature of 2 converter cabinets;

[0067] 6.3 Calculate the total error of the output layer, yi: actual value, Predicted value

[0068] Total error calculation

[0069]

[0070] 6.4 When the error does not meet the design goal, optimize until it meets the requirements, use the error back propagation algorithm and gradient descent method to solve and update the weights of the convolutional neural network model, update all filter values / weights and parameter values, iterate multiple times, and reduce the output error;

[0071] 6.5 Repeat steps 2-4 to complete the training of the neural network model.

[0072] In a further implementation, during the prediction phase, anomaly detection is performed by monitoring the difference between the predicted results and the actual observed values; when a temperature anomaly is detected, the locomotive system can automatically take measures, such as adjusting power system parameters, issuing alarms, or performing automatic maintenance.

[0073] Based on actual application conditions, the model is further optimized to improve prediction accuracy and real-time performance; the performance of system operation is monitored, and the model is continuously improved based on feedback to adapt to the needs of different environments and working conditions.

[0074] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for predicting the outlet water temperature of a locomotive converter cabinet based on a CNN model, characterized in that: The steps are as follows: S1: Collect historical data related to the outlet water temperature of the converter cabinet, pre-process the collected data to form a data set, and divide the data set into a training set, a validation set, and a test set; S2: Build CNN model; S3: Use the training set and validation set to train and validate the CNN model; S4: Use the test set to evaluate the trained CNN model; S5: The trained CNN model obtains real-time data related to the outlet water temperature of the converter cabinet, and uses the abnormal diagnosis algorithm based on real-time data in the CNN model to predict and diagnose faults.

2. The method for predicting the outlet water temperature of a locomotive converter cabinet based on a CNN model as claimed in claim 1, characterized in that: In step S1, preprocessing includes data cleaning, normalization and outlier processing.

3. The method for predicting the outlet water temperature of a locomotive converter cabinet based on a CNN model as claimed in claim 1, characterized in that: In step S2, the CNN model includes input layer 1, convolutional layers 1, 2, pooling layer 1, convolutional layer 3, pooling layer 2, flattening layer 1, and connection layers 1, 2.

4. The method for predicting the outlet water temperature of a locomotive converter cabinet based on a CNN model as claimed in claim 1, characterized in that: In step S5, the abnormality diagnosis algorithm calculation includes the following steps: 4.

1. If the predicted outlet water temperature of the converter cabinet is greater than the alarm temperature threshold, the CNN model determines that it is abnormal and an alarm is issued. Otherwise, the predicted outlet water temperature of the converter cabinet is compared with the actual outlet water temperature of the converter cabinet. 4.

2. If the difference between the predicted converter cabinet outlet water temperature and the actual converter cabinet outlet water temperature is greater than the threshold value, it is considered abnormal and an alarm is issued.

5. A method for predicting the outlet water temperature of a locomotive converter cabinet based on a CNN model as claimed in claim 4, characterized in that: the calculation of the threshold value The following steps are involved: 5.

1. Based on the test set, the CNN model is used to predict the data set, and the difference between the predicted data set and the test set is calculated as diff; 5.

2. Calculate the vector norm of diff and sort it; 5.

3. Use the preset cutoff value of the sorted vector norm as the threshold.

6. The method for predicting the outlet water temperature of a locomotive converter cabinet based on a CNN model as claimed in claim 1, characterized in that: In step S3, The following steps are involved: 6.

1. Initialize the neural network model, specify filters and parameters / weights; 6.

2. The data in the training set is used as input for training, and the output layer data can be obtained, namely, the outlet water temperature of 1 converter cabinet and the outlet water temperature of 2 converter cabinets; 6.

3. Calculate the total error of the output layer, yi: actual value, Predicted value Total error calculation 6.4 When the error does not meet the design goal, optimize until it meets the requirements, use the error back propagation algorithm and gradient descent method to solve and update the weights of the convolutional neural network model, update all filter values / weights and parameter values, iterate multiple times, and reduce the output error; 6.5 Repeat steps 2-4 to complete the training of the neural network model.

Images