Intelligent fault detection and diagnosis system for railway locomotive

Abstract

The invention relates to the technical field of railway locomotive maintenance, and discloses an intelligent fault detection and diagnosis system for a railway locomotive, which integrates three core units, namely a data preprocessing unit, a model construction unit and a fault analysis unit. Operation data are collected in real time through various sensors installed at key parts of a locomotive, and a rich training data set is formed by combining locomotive maintenance records and fault label data. A fault detection model and a fault diagnosis model are trained by using a machine learning algorithm, accurate detection of the abnormal state of the locomotive and rapid determination of the fault type are realized, and the specific reason and position of the fault are diagnosed at the same time. The system also innovatively constructs a residual error fitting model for correcting and optimizing fault detection and diagnosis results, and the diagnosis accuracy is improved. Finally, a detailed fault detection and diagnosis report is generated through the fault analysis unit, and powerful support is provided for timely maintenance and safety guarantee of the locomotive.

Description

Technical Field

[0001] The present invention relates to the technical field of railway locomotive maintenance, and in particular to an intelligent fault detection and diagnosis system for railway locomotives. Background Art

[0002] With the rapid development of the railway transportation industry, the safe, stable and efficient operation of railway locomotives as the core equipment of railway transportation is crucial to ensure the smooth operation of railway transportation. However, during the operation of railway locomotives, various faults will inevitably occur due to various factors such as mechanical wear, electrical failures, environmental factors, etc. If these faults are not discovered and handled in time, they will not only affect the normal operation of the locomotive, but may also cause serious safety accidents and cause huge losses to people's lives and property.

[0003] At present, the fault detection and diagnosis methods of railway locomotives mainly rely on manual inspection and regular maintenance. Manual inspection requires a lot of manpower and material resources, and is limited by the experience and skill level of the inspectors, making it difficult to detect faults comprehensively, accurately and promptly. Regular maintenance is often based on fixed maintenance cycles and maintenance standards, and cannot be flexibly adjusted according to the actual operating status of the locomotive, which can easily cause problems such as over-maintenance or under-maintenance. With the rapid development of information technology, especially the widespread application of technologies such as big data, cloud computing, and artificial intelligence, new technical means have been provided for the detection and diagnosis of railway locomotive faults. By collecting the operation data of the locomotive in real time and combining machine learning and deep learning algorithms, intelligent fault detection and diagnosis can be performed on the locomotive, realizing early detection and early processing of faults, and improving the operating safety and reliability of the locomotive. Summary of the invention

[0004] The object of the present invention is to provide a railway locomotive intelligent fault detection and diagnosis system to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a railway locomotive intelligent fault detection and diagnosis system, the system comprising:

[0006] The data preprocessing unit collects operation data from the actual operation of the railway locomotive, specifically through real-time collection of sensors installed at key parts of the locomotive, including speed sensors, temperature sensors, pressure sensors, vibration sensors, current sensors and voltage sensors; collects locomotive maintenance records, fault records and corresponding maintenance measures as fault label data; integrates and preprocesses the operation data and fault label data to form a training data set;

[0007] A model building unit is used to build a railway locomotive fault detection model and a fault diagnosis model, wherein the fault detection model is used to detect the abnormal state of the railway locomotive and determine the fault type, and the fault diagnosis model is used to diagnose the specific cause and location of the fault based on current and historical operating data; based on the training data set, the fault detection model and the fault diagnosis model are trained using a machine learning algorithm, and the residual fitting model is constructed using the residuals of the two models, and the residual fitting model is used to correct and optimize the results of fault detection and diagnosis;

[0008] The fault analysis unit uses the trained model and real-time locomotive operation data to generate a fault detection and diagnosis report for the railway locomotive.

[0009] Preferably, a support vector machine (SVM) algorithm is used to construct a fault detection model, and the high-dimensional feature space of the locomotive operation data is linearly divided through kernel function mapping to achieve detection and classification of abnormal conditions.

[0010] Preferably, the training step of the fault detection model includes:

[0011] Construct the support vector machine SVM structure: design the SVM model, select the radial basis function RBF as the kernel function, and set the penalty parameter C and kernel parameter γ;

[0012] Model training: Input the training data set into the SVM model and find the optimal hyperplane by solving the optimization problem;

[0013] Model verification and adjustment: Use the cross-validation method to evaluate the detection performance of the model, adjust the parameters C and γ according to the accuracy and recall rate indicators, and optimize the model structure.

[0014] Preferably, a random forest RF algorithm is used to construct a fault diagnosis model, and the specific cause and location of the fault are output through integrated learning of multiple decision trees.

[0015] Preferably, the training step of the fault diagnosis model includes:

[0016] Feature selection: Select key features related to faults from the training dataset;

[0017] Construct the random forest RF structure: define the RF model, determine the number of decision trees, the maximum depth, and the minimum number of sample splits;

[0018] Model training: Input feature data into the RF model, and obtain fault diagnosis results through the training and voting mechanism of multiple decision trees;

[0019] Model evaluation and optimization: Evaluate the diagnostic accuracy and generalization ability of the model on the test set, and adjust the parameters of the RF model based on the evaluation results, including increasing the number of decision trees or adjusting the depth of the tree.

[0020] Preferably, the step of constructing the residual fitting model includes:

[0021] Calculate the residuals between the actual outputs of the fault detection model and the fault diagnosis model and the true labels, which are recorded as detection residuals and diagnosis residuals respectively;

[0022] Combining the detection residuals and the diagnosis residuals with the original operation data to form a residual feature set;

[0023] The gradient boosting tree (GBDT) algorithm is used to train the residual feature set and construct the residual fitting model;

[0024] The structure of the residual fitting model includes an input layer, an integration layer composed of multiple decision trees and an output layer. The input layer receives the residual feature set, and the output layer outputs the corrected fault detection and diagnosis results, and the residual is reduced through the stepwise fitting of GBDT.

[0025] Preferably, the data preprocessing unit preprocesses the collected data in a manner including data cleaning, encoding processing and standardization processing.

[0026] Preferably, the method of cleaning the data includes:

[0027] For missing values, linear interpolation algorithm is used to fill them;

[0028] For outliers, the threshold judgment method is used to regard data points that exceed the preset threshold range as outliers and replace them with the mean of their neighboring points.

[0029] Preferably, the method of encoding the data includes: performing one-hot encoding conversion on discrete features.

[0030] Preferably, the method for standardizing the data includes: using the Z-score standardization method to standardize the continuous features, specifically: subtracting the mean of each numerical data and dividing it by its standard deviation, so that the processed data conforms to the standard normal distribution, and the standardization formula is: Z = (X-μ) / σ, wherein X is the original data, μ is the mean of the original data, μ is the standard deviation of the original data, and Z is the standardized data.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] This system collects real-time operation data, including speed, temperature, pressure, vibration, current, voltage, etc., through a variety of sensors installed in key parts of the locomotive, ensuring the comprehensiveness and real-time nature of the data. Combined with machine learning algorithms, the fault detection model can accurately identify the abnormal state of the locomotive and determine the type of fault, greatly improving the accuracy and timeliness of fault detection, and effectively avoiding safety accidents caused by failure to discover the fault in time. Based on current and historical operation data, the fault diagnosis model can deeply explore the correlation between faults and operation data, so as to accurately diagnose the specific cause and location of the fault. This not only helps to quickly formulate effective maintenance measures, but also greatly shortens the fault handling time and improves the operation efficiency of the locomotive.

[0033] This system introduces a residual fitting model and uses the residuals of the fault detection model and fault diagnosis model for correction and optimization, which effectively reduces the error in the model prediction process and further improves the accuracy and reliability of fault detection and diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a working principle diagram of a railway locomotive intelligent fault detection and diagnosis system according to the present invention;

[0035] Figure 2 This is the training flow chart of the fault detection model;

[0036] Figure 3 This is the training flow chart of the fault diagnosis model;

[0037] Figure 4 Training flowchart for fitting a model to residuals. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of 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.

[0039] See also Figure 1-4 The present invention provides a technical solution: a railway locomotive intelligent fault detection and diagnosis system, the system comprising:

[0040] Data preprocessing unit: The data preprocessing unit is responsible for collecting operation data from the actual operation of the railway locomotive. Specifically, by installing sensors such as speed sensors, temperature sensors, pressure sensors, vibration sensors, current sensors and voltage sensors at key parts of the locomotive, the locomotive's operating status information is collected in real time. At the same time, the locomotive's maintenance records, fault records and corresponding maintenance measures are collected as fault label data. The data is integrated and preprocessed to form a high-quality training data set.

[0041] Model building unit: The model building unit is the core part of this system, which is mainly responsible for building railway locomotive fault detection model and fault diagnosis model.

[0042] (1) Fault detection model: This model uses machine learning algorithms to train preprocessed operating data to learn the normal operating mode and abnormal state characteristics of the locomotive. After the model training is completed, it can detect the operating status of the locomotive in real time, determine whether there is an abnormality, and preliminarily determine the fault type.

[0043] (2) Fault diagnosis model: Based on the fault detection model, the fault diagnosis model further uses deep learning methods to conduct in-depth analysis of the current and historical operating data of the locomotive to diagnose the specific cause and location of the fault. This model can combine the operating rules and fault mechanisms of the locomotive to provide more accurate fault diagnosis results.

[0044] (3) Residual fitting model: In order to further improve the accuracy of fault detection and diagnosis, this system also constructs a residual fitting model. This model uses the residual information of the fault detection model and the fault diagnosis model to correct and optimize the prediction results of the original model by fitting the residual data, thereby obtaining more reliable fault detection and diagnosis results.

[0045] Fault Analysis Unit: The fault analysis unit is responsible for using the trained model and real-time locomotive operation data to generate a fault detection and diagnosis report for the railway locomotive. The report will list the locomotive fault conditions in detail, including fault type, fault cause, fault location, and recommended repair measures, etc., providing a scientific basis for locomotive repair and maintenance.

[0046] The present invention will be further described below in conjunction with Examples 1 to 4:

[0047] Embodiment 1:

[0048] In order to realize the fault detection of the locomotive running state, this embodiment adopts the support vector machine (SVM) algorithm to build a fault detection model, and uses the kernel function mapping technology to linearly divide the high-dimensional feature space of the locomotive running data to effectively detect and classify abnormal states. The following are the specific training steps and implementation details of the fault detection model:

[0049] ①Build support vector machine SVM structure:

[0050] Design the SVM model and select the radial basis function (RBF) as the kernel function. The RBF kernel function can map the original data to a high-dimensional space, making the data in this high-dimensional space linearly separable. The expression of the RBF kernel function is:

[0051] K(xi|xj)=exp(-γ‖xi-xj‖ 2 )

[0052] Among them, xi and xj are two sample points, and γ is the kernel parameter, which determines the distribution complexity after data mapping.

[0053] Set the penalty parameter C. Parameter C is used to control the degree of penalty for misclassification. The larger the C value, the heavier the penalty for misclassification and the more likely the model is to overfit. The smaller the C value, the more likely the model is to underfit. Therefore, a reasonable choice of C value is crucial to model performance.

[0054] ②Model training:

[0055] The preprocessed training data set is input into the SVM model. The training data set includes locomotive operation data in normal and abnormal states, and each data sample contains multiple features, such as speed, temperature, pressure, etc.

[0056] The training process of the SVM model is essentially to solve an optimization problem, that is, to find a hyperplane that can maximize the margin while minimizing the classification error. Under the RBF kernel function, this optimization problem can be solved in the Lagrangian dual form, and algorithms such as Sequential Minimal Optimization (SMO) can be used to efficiently find the optimal solution.

[0057] After the training is completed, a set of support vectors and parameters of the optimal hyperplane are obtained. These parameters define the decision boundary for subsequent data classification.

[0058] ③Model verification and adjustment:

[0059] In order to evaluate the detection performance of the model, the cross-validation method is used. Specifically, the training data set is divided into several subsets, one of which is used as the validation set each time, and the remaining subsets are used as the training set, and multiple training and validation are performed. The performance of the model is evaluated by calculating the accuracy and recall of each validation.

[0060] Accuracy: The ratio of correctly classified samples to the total number of samples.

[0061] Recall rate: The ratio of samples correctly identified as abnormal to all samples with abnormal status.

[0062] According to the results of cross-validation, adjust the penalty parameter C and the kernel parameter γ. You can use methods such as grid search (GridSearch) or random search (Random Search) to find the optimal combination within the preset parameter range. After each parameter adjustment, retrain the model and perform cross-validation until you find the best performance parameter combination.

[0063] Finally, the optimized model is used to detect faults on the test data set to verify the effectiveness of the model in practical applications. If the model performance meets the requirements, the model can be deployed in the locomotive operation status monitoring system to achieve real-time detection and classification of locomotive faults.

[0064] Embodiment 2:

[0065] To achieve efficient and accurate fault diagnosis, this embodiment uses the Random Forest (RF) algorithm to build a fault diagnosis model. The model integrates multiple decision trees and uses the voting mechanism of each tree to output the specific cause and location of the fault. The following are the specific training steps and implementation details of the fault diagnosis model:

[0066] ① Feature selection:

[0067] Select key features that are closely related to the fault from the original training data set. This process can be achieved through feature importance evaluation methods, such as using the Gini Index, Information Gain, or model-based feature selection methods (such as using linear models to evaluate feature weights). The selected features should be able to comprehensively and effectively reflect the fault status while avoiding the introduction of redundant and irrelevant information.

[0068] ②Build a random forest RF structure:

[0069] Define the RF model and determine its key parameters, including the number of decision trees, the maximum depth of each tree, and the minimum number of sample splits.

[0070] Number of decision trees: Determines the total number of trees in the RF model, usually determined empirically or through cross-validation to ensure that the model performance is stable and not overfitting.

[0071] Maximum depth: Limit the maximum growth depth of each tree to prevent overfitting caused by too deep a tree. You can stop by setting the maximum depth to a fixed value or allowing the tree to grow until the number of samples contained in the leaf node is less than the minimum number of sample splits.

[0072] Minimum number of sample splits: Determines the minimum number of samples required for a tree node to split, which helps control the complexity of the tree and the risk of overfitting.

[0073] ③Model training:

[0074] The data after feature selection is input into the RF model for training. The RF model achieves classification or regression of data by constructing multiple decision trees and randomly selecting features for splitting at the nodes of each tree. During the training process, each tree learns the data independently and obtains the final diagnosis result through a voting mechanism.

[0075] Specifically, for each sample, the RF model traverses all decision trees and records the prediction results of each tree for the sample. For classification problems, the final prediction result is the category voted by the majority of trees; for regression problems, it is the average of the prediction values ​​of each tree.

[0076] ④Model evaluation and optimization:

[0077] Evaluate the diagnostic accuracy and generalization ability of the RF model on an independent test set. Evaluation metrics may include precision, recall, F1 score, etc. Based on the evaluation results, adjust the parameters of the RF model to optimize performance. For example, if the model performs poorly on the test set, consider increasing the number of decision trees, adjusting the maximum depth of the tree, or the minimum number of sample splits to improve the generalization ability and diagnostic accuracy of the model.

[0078] Additionally, strategies such as cross-validation, grid search, or random search can be used to systematically explore the parameter space and find the model configuration with the best performance.

[0079] Embodiment 3:

[0080] In order to further improve the accuracy of fault detection and diagnosis, this embodiment proposes a method for constructing a residual fitting model. The method calculates the residuals between the actual outputs of the fault detection model and the fault diagnosis model and the true labels, and fits these residuals using the Gradient Boosting Decision Tree (GBDT) algorithm, thereby correcting the original model output. The following are the specific implementation steps:

[0081] ① Calculate the residual:

[0082] For the existing fault detection model and fault diagnosis model, the difference between their output on the actual operation data and the true label is calculated to obtain the detection residual and diagnosis residual. The detection residual reflects the error of the fault detection model in judging whether there is a fault, while the diagnosis residual reflects the error of the fault diagnosis model in determining the fault type or location.

[0083] Specifically, for each sample data point, its detection residual is calculated as the difference between the output of the fault detection model and the label of whether the real fault exists; its diagnosis residual is calculated as the difference between the output of the fault diagnosis model and the label of the real fault type or location.

[0084] ②Forming residual feature set:

[0085] The calculated detection residuals and diagnostic residuals are combined with the original operation data to form a new feature set, namely the residual feature set. This feature set not only contains the original operation data features, but also contains the residual features that reflect the model errors, providing rich information for subsequent residual fitting.

[0086] For example, if the original operating data includes features such as speed, temperature, and pressure, the residual feature set will contain these features and the corresponding detection residuals and diagnosis residuals.

[0087] ③Construct residual fitting model:

[0088] The GBDT algorithm is used to train the residual feature set and construct the residual fitting model. GBDT is a powerful ensemble learning method that approximates complex functions by gradually training multiple decision trees and weighting their prediction results.

[0089] The structure of the residual fitting model includes an input layer, an integration layer consisting of multiple decision trees, and an output layer. The input layer receives the residual feature set as input; the integration layer consists of multiple decision trees, each of which splits and predicts the input features; and the output layer outputs the corrected fault detection and diagnosis results.

[0090] During the training process, the GBDT algorithm gradually fits the mapping relationship between the residual feature set and the true label, and reduces the residual and improves the prediction accuracy by continuously adjusting the parameters and structure of the decision tree.

[0091] ④Model training and optimization:

[0092] The residual feature set is divided into a training set and a validation set. The GBDT model is trained using the training set, and the performance of the model is evaluated using the validation set. Based on the evaluation results, the parameters of the GBDT model, such as the number, depth, and learning rate of decision trees, can be adjusted to further optimize the model performance.

[0093] Through continuous training and optimization, the residual fitting model can gradually learn the error patterns in the fault detection and diagnosis model and correct these errors. Ultimately, combining the residual fitting model with the original fault detection and diagnosis model can achieve more accurate detection and diagnosis of faults.

[0094] Embodiment 4:

[0095] In order to ensure the quality and consistency of data, and thus improve the accuracy of subsequent fault detection and diagnosis models, this embodiment describes in detail a method for data preprocessing. The method includes three steps: data cleaning, encoding processing, and standardization processing. The following are specific implementation details:

[0096] ①Data cleaning:

[0097] Data cleaning aims to eliminate missing values ​​and outliers in the data set to ensure the integrity and reliability of the data.

[0098] Missing value processing: For missing values ​​in the data set, a linear interpolation algorithm is used to fill them. Specifically, for missing points in time series data, the values ​​of the two known data points before and after are used to calculate the estimated value of the missing point through a linear interpolation formula, and this is used to fill the missing value. This method can maintain the continuity and trend of the data.

[0099] Outlier processing: The threshold judgment method is used to identify and process outliers. First, according to the statistical characteristics of the data and business logic, a reasonable threshold range is set for each feature. Then, each data point in the data set is traversed, and if its value exceeds the preset threshold range, it is considered an outlier. For the identified outliers, the mean of its neighboring points is used to replace them to eliminate their interference with model training.

[0100] ②Encoding processing:

[0101] The encoding process is mainly used to convert discrete features into numerical forms that the model can understand. In this embodiment, one-hot encoding is used to convert discrete features. Specifically, for each discrete feature, a binary vector of equal length is created according to the number of possible values. For each sample, the corresponding feature value position is marked as 1, and the remaining positions are marked as 0. In this way, each discrete feature is converted into a unique binary vector, which is convenient for subsequent processing and learning of the model.

[0102] ③ Standardization processing:

[0103] The purpose of standardization is to eliminate the dimensional differences between different features and make the data comparable. In this embodiment, the Z-score standardization method is used to standardize the continuous features. Specifically, for each continuous feature, its mean and standard deviation are calculated. Then, each numerical data is subtracted from its mean and divided by its standard deviation to obtain the standardized data. The formula for standardization is as follows:

[0104] Z=(X-μ) / σ

[0105] Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data. Through standardization, the processed data can conform to the standard normal distribution, which is conducive to the training and convergence of the model.

[0106] Taking the operating data of a certain device as an example, assume that the data set contains continuous features such as speed, temperature, and pressure, as well as discrete features such as equipment type and operating status. First, clean the missing values ​​and outliers in the data set to ensure the integrity and accuracy of the data. Then, perform one-hot encoding conversion on discrete features such as equipment type and operating status to convert them into a numerical form that the model can understand. Finally, perform Z-score standardization on continuous features such as speed, temperature, and pressure to eliminate dimensional differences. After the above preprocessing steps, the quality and consistency of the data set have been significantly improved.

[0107] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0108] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A railway locomotive intelligent fault detection and diagnosis system, characterized in that: The system comprises: The data preprocessing unit collects operation data from the actual operation of the railway locomotive, specifically through real-time collection of sensors installed at key parts of the locomotive, including speed sensors, temperature sensors, pressure sensors, vibration sensors, current sensors and voltage sensors; collects locomotive maintenance records, fault records and corresponding maintenance measures as fault label data; integrates and preprocesses the operation data and fault label data to form a training data set; A model building unit is used to build a railway locomotive fault detection model and a fault diagnosis model, wherein the fault detection model is used to detect the abnormal state of the railway locomotive and determine the fault type, and the fault diagnosis model is used to diagnose the specific cause and location of the fault based on current and historical operating data; based on the training data set, the fault detection model and the fault diagnosis model are trained using a machine learning algorithm, and the residual fitting model is constructed using the residuals of the two models, and the residual fitting model is used to correct and optimize the results of fault detection and diagnosis; The fault analysis unit uses the trained model and real-time locomotive operation data to generate a fault detection and diagnosis report for the railway locomotive.

2. The intelligent fault detection and diagnosis system for railway locomotives according to claim 1, characterized in that: The support vector machine (SVM) algorithm is used to build a fault detection model. The high-dimensional feature space of locomotive operation data is linearly divided through kernel function mapping to achieve detection and classification of abnormal conditions.

3. A railway locomotive intelligent fault detection and diagnosis system according to claim 2, characterized in that: The training steps of the fault detection model include: Construct the support vector machine SVM structure: design the SVM model, select the radial basis function RBF as the kernel function, and set the penalty parameter C and kernel parameter γ; Model training: Input the training data set into the SVM model and find the optimal hyperplane by solving the optimization problem; Model verification and adjustment: Use the cross-validation method to evaluate the detection performance of the model, adjust the parameters C and γ according to the accuracy and recall rate indicators, and optimize the model structure.

4. The intelligent fault detection and diagnosis system for railway locomotives according to claim 1, characterized in that: The random forest RF algorithm is used to build a fault diagnosis model, and the specific cause and location of the fault are output through ensemble learning of multiple decision trees.

5. The railway locomotive intelligent fault detection and diagnosis system according to claim 1, characterized in that: The training steps of the fault diagnosis model include: Feature selection: Select key features related to faults from the training dataset; Construct the random forest RF structure: define the RF model, determine the number of decision trees, the maximum depth, and the minimum number of sample splits; Model training: Input feature data into the RF model, and obtain fault diagnosis results through the training and voting mechanism of multiple decision trees; Model evaluation and optimization: Evaluate the diagnostic accuracy and generalization ability of the model on the test set, and adjust the parameters of the RF model based on the evaluation results, including increasing the number of decision trees or adjusting the depth of the tree.

6. A railway locomotive intelligent fault detection and diagnosis system according to claim 1, characterized in that: The steps to construct a residual fitting model include: Calculate the residuals between the actual outputs of the fault detection model and the fault diagnosis model and the true labels, which are recorded as detection residuals and diagnosis residuals respectively; Combining the detection residuals and the diagnosis residuals with the original operation data to form a residual feature set; The gradient boosting tree (GBDT) algorithm is used to train the residual feature set and construct the residual fitting model; The structure of the residual fitting model includes an input layer, an integration layer composed of multiple decision trees and an output layer. The input layer receives the residual feature set, and the output layer outputs the corrected fault detection and diagnosis results, and the residual is reduced through the stepwise fitting of GBDT.

7. The railway locomotive intelligent fault detection and diagnosis system according to claim 1, characterized in that: The data preprocessing unit preprocesses the collected data in the following ways: data cleaning, encoding and standardization.

8. A railway locomotive intelligent fault detection and diagnosis system according to claim 7, characterized in that: Methods for cleaning data include: For missing values, linear interpolation algorithm is used to fill them; For outliers, the threshold judgment method is used to regard data points that exceed the preset threshold range as outliers and replace them with the mean of their neighboring points.

9. A railway locomotive intelligent fault detection and diagnosis system according to claim 7, characterized in that: Methods for encoding data include: one-hot encoding conversion of discrete features.

10. The railway locomotive intelligent fault detection and diagnosis system according to claim 7, characterized in that: The method of standardizing the data includes: using the Z-score standardization method to standardize the continuous features, specifically: subtracting the mean of each numerical data and dividing it by its standard deviation, so that the processed data conforms to the standard normal distribution. The standardization formula is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, μ is the standard deviation of the original data, and Z is the standardized data.

Images