Freight railway empty car matching method and system based on gradient boosting decision tree algorithm
By constructing a railway empty car matching model using the gradient boosting decision tree algorithm, the problems of adaptability and portability in empty car allocation are solved, achieving low-cost and efficient empty car matching, which is applicable to multiple business scenarios and railway lines.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CRSC RESEARCH & DESIGN INSTITUTE GROUP CO LTD
- Filing Date
- 2026-03-09
- Publication Date
- 2026-07-14
Smart Images
Figure CN122390250A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of railway freight technology, and in particular to a method and system for matching empty freight trains based on gradient boosting decision tree algorithm. Background Technology
[0002] Loading and unloading operations in heavy-haul railway freight transport are divided into unloading and loading ends. Loaded cars are unloaded at the unloading end and become empty cars, which are then returned to the loading end for loading. To ensure the implementation of loading plans and improve transport efficiency, it is necessary to rationally allocate empty car resources and promptly dispatch them to suitable loading stations. However, based on current freight operations and dispatchers' work habits, it is impossible to determine the matching relationship between daily freight plans and empty cars based on information such as car type, cargo category, loading location, and car ownership, making it difficult to select usable empty cars.
[0003] Existing solutions to the empty car allocation problem in freight railways mostly involve establishing mathematical models and solving these models to obtain the optimal empty car allocation plan. These mathematical models are primarily based on operations research and graph theory, constructing optimization models such as linear programming and multi-objective programming. However, existing solutions have the following drawbacks: 1. The model mainly considers meeting the empty car demand at loading stations and reducing empty car transportation time and distance. It lacks research on the matching relationship between the specific characteristics of empty cars and the daily freight plan demand, such as the characteristics of empty car type and cargo type.
[0004] 2. The model has significant limitations; it cannot be directly migrated and applied across different business scenarios or railway lines, and it requires repeated updates when new business requirements are proposed.
[0005] 3. The application cost is high and the development cycle is long. In actual work, it is difficult to build and apply the model in a targeted manner. Most of the time, the dispatching of empty vehicles is done manually by the dispatcher. Summary of the Invention
[0006] To address the aforementioned issues, this application provides a method and system for matching empty freight trains based on a gradient boosting decision tree algorithm. This method can match suitable empty trains according to the needs of the daily freight schedule, achieving high adaptability and portability while reducing the development cycle and application costs.
[0007] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, the present invention provides a method for matching empty freight cars on railways based on a gradient boosting decision tree algorithm, comprising: Historical data on empty car matching for freight railways are preprocessed to obtain positive sample data for the empty car matching model of freight railways. The empty car data that were not matched in each daily plan in the positive sample data are concatenated with the daily plan to obtain the negative sample data of the freight railway empty car matching model; The positive and negative sample data of the freight railway empty car matching model are merged to obtain the target training set of the freight railway empty car matching model. Based on the target training set, the gradient boosting decision tree model is trained using the k-fold cross-validation method to obtain the freight railway empty car matching model; The given daily schedule and empty car data are input into the freight railway empty car matching model for matching and prediction, and the allocation results of the daily schedule and empty cars are obtained.
[0008] Furthermore, the historical data on empty car matching for freight railways is preprocessed, including: Remove null values from the historical data of empty freight railway car matching. Label and encode the category features carried by the textual information in the historical data of empty freight railway car matching; Convert discrete attributes from historical data on empty car matching for freight railways into numerical values that can be used by the model; Specific information from the historical data of empty car matching on freight railways is identified as new features.
[0009] Furthermore, based on the target training set, the gradient boosting decision tree model is trained using the k-fold cross-validation method to obtain a freight railway empty car matching model, including: Based on the optimized parameters of sample weights and learning rate, a basic gradient boosting decision tree model is constructed. Based on the target training set, the k-fold cross-validation method is used to perform k iterations of training and result validation on the basic gradient boosting decision tree model. The basic gradient boosting decision tree model with the best average performance among the validation results of k iterations of training is determined as the empty car matching model for freight railways.
[0010] Furthermore, based on the target training set, the basic gradient boosting decision tree model is trained and validated k times using a k-fold cross-validation method, including: The target training set is randomly divided into k subsets of equal size. In each iteration, one subset is selected as the validation set, and the remaining k-1 subsets are selected as the sub-training set. The basic gradient boosting decision tree model is trained k times based on the sub-training set; The k-th iteration training results of the basic gradient boosting decision tree model are validated based on the validation set corresponding to the sub-training set.
[0011] Secondly, the present invention also provides a freight railway empty car matching system based on gradient boosting decision tree algorithm, comprising: The processing module is used to preprocess the historical data of empty car matching for freight railways to obtain positive sample data for the empty car matching model of freight railways. The splicing module is used to splice the empty car data that were not matched in each daily plan in the positive sample data with the daily plan to obtain the negative sample data of the freight railway empty car matching model; The merging module is used to merge the positive and negative sample data of the freight railway empty car matching model to obtain the target training set of the freight railway empty car matching model. The training module is used to train the gradient boosting decision tree model based on the target training set using the k-fold cross-validation method, so as to obtain the freight railway empty car matching model. The prediction module is used to input the given daily plan and empty car data into the freight railway empty car matching model for matching and prediction, and obtain the daily plan and empty car allocation results.
[0012] Furthermore, the historical data on empty car matching for freight railways is preprocessed, including: Remove null values from the historical data of empty freight railway car matching. Label and encode the category features carried by the textual information in the historical data of empty freight railway car matching; Convert discrete attributes from historical data on empty car matching for freight railways into numerical values that can be used by the model; Specific information from the historical data of empty car matching on freight railways is identified as new features.
[0013] Furthermore, the training module includes: The building unit is used to optimize parameters based on sample weights and learning rate to build a basic gradient boosting decision tree model; The iterative unit is used to perform k iterations of training and result verification on the basic gradient boosting decision tree model based on the target training set using the k-fold cross-validation method. The unit is used to identify the best-performing base gradient boosting decision tree model from the validation results of k iterations of training as the empty car matching model for freight railways.
[0014] Furthermore, based on the target training set, the basic gradient boosting decision tree model is trained and validated k times using a k-fold cross-validation method, including: The target training set is randomly divided into k subsets of equal size. In each iteration, one subset is selected as the validation set, and the remaining k-1 subsets are selected as the sub-training set. The basic gradient boosting decision tree model is trained k times based on the sub-training set; The k-th iteration training results of the basic gradient boosting decision tree model are validated based on the validation set corresponding to the sub-training set.
[0015] Thirdly, the present invention also provides an electronic device, comprising: a processor and a memory; The processor is coupled with the memory; The processor is used to read and execute programs or instructions stored in the memory, causing the device to perform the method as described in the first aspect.
[0016] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method as described in the first aspect.
[0017] The technical solution provided in this application has at least the following technical effects or advantages: The technical solution of this application performs feature engineering on the daily plans and empty car information in the historical data of freight railway empty car matching, extracting specific features of both, and focusing more on the relationship between these specific features and the matching possibility. A freight railway empty car matching model is constructed based on the gradient boosting decision tree algorithm in machine learning to solve the freight railway empty car matching problem. This model has good scalability, adaptability, and portability, and can be quickly migrated and applied across different business scenarios or railway lines. Furthermore, the model can be quickly updated when new business requirements arise. The technical solution of this application also has the advantages of low application cost and short development cycle.
[0018] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 This is a flowchart illustrating a method for matching empty freight train cars based on a gradient boosting decision tree algorithm, as described in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of a freight railway empty car matching system based on the gradient boosting decision tree algorithm in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0022] Daily freight railway empty car matching generally falls into several categories (e.g., integrated, non-integrated, non-coal). When submitting the daily plan, information such as car type, cargo category, loading location, and car ownership is specified. Empty small train numbers or empty 10,000-ton trains also have characteristics such as car type, ministry / enterprise markings, and train number. However, there is no absolute one-to-one correspondence between the characteristics of the daily plan and the characteristics of the empty cars. For example, common C64 and C70 small trains or 10,000-ton trains can be used in both non-coal and integrated daily plans. This makes it difficult to select usable empty cars when matching empty cars for a specific type of daily plan.
[0023] This application is based on the gradient boosting decision tree algorithm in machine learning. It uses business data of freight railway empty car matching to train the model. Then, the model can be used to match the daily plan and empty cars of freight railway empty car matching, and determine whether the empty car can be assigned to the daily plan for empty car allocation. In this way, it can match suitable empty cars for empty car allocation according to the needs of the daily freight plan, and achieve the purpose of high adaptability and portability, reducing the development cycle and application cost.
[0024] Machine learning refers to the theories and methods by which computers acquire knowledge and skills by analyzing data and simulating human learning activities, thereby improving system performance. The process includes building an algorithm model, training the algorithm model with data (analyzing patterns from the data), and using the trained algorithm to complete tasks (based on the learning patterns, it is divided into classifying and predicting unknown data).
[0025] Gradient boosting decision trees are a machine learning algorithm that improves model performance by building and combining multiple machine learning models. The basic principle is to iteratively train the decision tree, optimizing it with each training iteration based on the results of previous iterations. By training multiple models and combining their predictions in some way, better prediction results can often be obtained than with a single model. This algorithm exhibits high performance in classification and regression tasks.
[0026] Figure 1 This is a flowchart illustrating a method for matching empty freight cars on a railway based on a gradient boosting decision tree algorithm, as described in this application. The method includes: S101. Preprocess the historical data of empty car matching for freight railways to obtain positive sample data for the empty car matching model of freight railways. Based on the scale of business data and business needs, historical data on empty freight railway car matching is retrieved from the database. This includes daily freight plans that have already been matched with empty cars and specific information about the corresponding empty cars, such as the plan submission time, loading station, destination station, cargo type, number of cars required, car type required, car ownership, as well as empty car number, car type, originating station, number of cars, and car ownership. Specific data information can be added to the retrieved historical data according to the specific business requirements. The more detailed the information and the more distinct the features, the better the training effect for subsequent models.
[0027] Preprocessing of historical data on empty car matching for freight railways includes null value removal and feature engineering. Since the data contains a significant amount of categorical information, label encoding can be performed on the categorical features carried by the textual information in the historical data, converting the discrete attributes into numerical representations usable by the model. Furthermore, specific information in the historical data can be identified as new features; that is, extracting special information from the historical data to establish new features, such as whether the car type is open wagon, whether it is a container wagon, the number of digits in the car number, the format of the car number, etc.
[0028] Existing solutions mostly employ the method of adding constraints to reflect the information features carried by the data into the model. This approach requires significant work when new categories are added to the business data features, necessitating model adjustments and the addition of new constraints. In contrast, the feature engineering approach used in this application does not require major model modifications. Label encoding automatically encodes new categories, allowing for model retraining and updates.
[0029] Since the historical data on empty car matching for freight railways only includes daily plans and corresponding allocated empty car data, the preprocessed historical data is used as positive sample data for training the empty car matching model for freight railways.
[0030] S102. The empty car data that were not matched in each daily plan in the positive sample data are concatenated with the daily plan to obtain the negative sample data of the freight railway empty car matching model. Negative sample data refers to the data in the historical data of freight railway empty car matching where daily plans and empty cars do not match. Negative sample data is unavailable and must be generated manually. The method for generating negative sample data involves concatenating daily plans and unmatched empty cars from the historical data of freight railway empty car matching. Specifically, for each daily plan in the historical data, 10 empty cars that are not matched with that daily plan are randomly selected from the integrated, non-integrated, and non-coal empty car matching records for that day. The daily plan and these empty cars are then concatenated into a single record, which becomes the negative sample data for the freight railway empty car matching model. The advantage of this technical solution for generating negative samples is that it starts from actual business data records, enabling the discovery of implicit business logic in actual empty car matching, which can then be used to train the model. Existing technical solutions are limited to imposing constraints on the model; when the constraints proposed by the business are insufficient, the model cannot constrain the implicit business logic.
[0031] S103. Merge the positive and negative sample data of the freight railway empty car matching model to obtain the target training set of the freight railway empty car matching model. The positive and negative sample data of the freight railway empty car matching model are merged, shuffled, and split into datasets to obtain the target training set and test set for the freight railway empty car matching model, which can be used for training and performance evaluation.
[0032] S104. Based on the target training set, the gradient boosting decision tree model is trained using the k-fold cross-validation method to obtain the freight railway empty car matching model. Based on the optimized parameters of sample weights and learning rate, a basic gradient boosting decision tree model is constructed. Based on the target training set, the basic gradient boosting decision tree model is trained and validated for k iterations using the k-fold cross-validation method. The basic gradient boosting decision tree model with the best average performance among the validation results of k iterations is determined as the empty car matching model for freight railways.
[0033] Specifically, the XGBClassifier method from the xgboost library in Python is used to construct a basic gradient boosting decision tree model. The gradient boosting decision tree model integrates many decision tree models to form a powerful classifier. This involves multiple trees making decisions together, and the result of each tree is the difference between the target value and the predictions of all previous trees. All results are then summed to obtain the final result, thereby improving the overall model performance.
[0034] When constructing and training the basic gradient boosting decision tree model, multiple iterative optimizations are performed on the traditional GBDT model based on parameters such as sample weights, learning rate, leaf node weights, and objective function. (1) In the daily schedule and empty vehicle matching problem, there may be a class imbalance problem between positive and negative samples. This can be optimized by weighting the loss function, giving higher weights to minority class samples in the loss function:
[0035] in, It is the weight of sample i, which is usually inversely proportional to the class frequency; This represents the total weighted loss value; n represents the number of samples in the training set. This represents the true value of sample i; This represents the predicted value for sample i; This represents the loss function value for sample i.
[0036] (2) An optimization method based on gradient descent is introduced, as shown in the following formula:
[0037] in, It is the weight of the j-th iteration. L is the learning rate, and L is the loss function. It represents a partial differential.
[0038] The learning rate can be dynamically adjusted based on changes in loss during model training, as shown in the following formula:
[0039] in, It is the learning rate in the j-th iteration. It is the initial learning rate. It is the attenuation coefficient.
[0040] (3) To prevent overfitting, L2 regularization can be added to the objective function:
[0041] in, This represents the total loss value after adding L2 regularization, where L is the original loss function and λ is the regularization coefficient. It is the m-th parameter of the model, where n represents the number of model parameters.
[0042] (4) Leaf node weight penalty: In the decision tree, the weight of the leaf nodes is penalized:
[0043] in, Let γ represent the complexity penalty term for the t-th tree, where γ is the penalty coefficient and T is the number of leaf nodes in the t-th tree. λ is the weight of the k-th leaf node, and λ is the regularization coefficient.
[0044] (5) In order to adapt to the dynamic changes in business data, an online gradient update learning method is introduced to gradually update the model.
[0045]
[0046] in, These are the model parameters in the j-th iteration. It's the learning rate. It is the gradient of the loss function.
[0047] Using the target training set, a k-fold cross-validation method is employed to train and validate the basic gradient boosting decision tree model for k iterations. Specifically, the target training set is randomly divided into k equal subsets. In each iteration, one subset is selected as the validation set, and the remaining k-1 subsets are used as sub-training sets. The basic gradient boosting decision tree model is trained k times based on these sub-training sets. The results of these k iterations are then validated using the corresponding validation sets. The basic gradient boosting decision tree model with the best average performance among the validation results from the k iterations is selected as the freight railway empty car matching model. The advantages of this model are: good scalability (updating the model without extensive modifications due to new business needs, requiring only training on the updated dataset); good portability (minor modifications to data processing and model parameters are sufficient for migration between different companies or railway bureaus' freight railway operations); and low application cost (requiring less work and computing power for model construction and training).
[0048] S105. Input the given daily plan and empty car data into the freight railway empty car matching model for matching and prediction to obtain the daily plan and empty car allocation results.
[0049] Given a daily schedule and relevant features of an empty wagon, after data preprocessing and feature engineering, the data is input into the freight railway empty wagon matching model to predict whether the two can match. The output of the freight railway empty wagon matching model can be a direct result of whether they match, or it can be a probability representing the degree of matching that the empty wagon can be assigned to the daily schedule. This allows for the rapid screening of qualified empty wagons and their allocation to the daily schedule for subsequent detailed empty wagon dispatch planning.
[0050] Furthermore, the feature importance and decision-making process of the freight railway empty car matching model can be visualized to more clearly explain the research on daily planning and empty car matching processes.
[0051] The technical solutions in this application have at least the following technical effects or advantages: The technical solution of this application performs feature engineering on the daily plans and empty car information in the historical data of freight railway empty car matching, extracting specific features of both, and focusing more on the relationship between these specific features and the matching possibility. A freight railway empty car matching model is constructed based on the gradient boosting decision tree algorithm in machine learning to solve the freight railway empty car matching problem. This model has good scalability, adaptability, and portability, and can be quickly migrated and applied across different business scenarios or railway lines. Furthermore, the model can be quickly updated when new business requirements arise. The technical solution of this application also has the advantages of low application cost and short development cycle.
[0052] Figure 2 This is a schematic diagram of a freight railway empty car matching system based on gradient boosting decision tree algorithm provided in this application embodiment. As shown in the figure, the system includes: The processing module is used to preprocess the historical data of empty car matching for freight railways to obtain positive sample data for the empty car matching model of freight railways. The splicing module is used to splice the empty car data that were not matched in each daily plan in the positive sample data with the daily plan to obtain the negative sample data of the freight railway empty car matching model; The merging module is used to merge the positive and negative sample data of the freight railway empty car matching model to obtain the target training set of the freight railway empty car matching model. The training module is used to train the gradient boosting decision tree model based on the target training set using the k-fold cross-validation method, so as to obtain the freight railway empty car matching model. The prediction module is used to input the given daily plan and empty car data into the freight railway empty car matching model for matching and prediction, and obtain the daily plan and empty car allocation results.
[0053] Furthermore, the historical data on empty car matching for freight railways is preprocessed, including: Remove null values from the historical data of empty freight railway car matching. Label and encode the category features carried by the textual information in the historical data of empty freight railway car matching; Convert discrete attributes from historical data on empty car matching for freight railways into numerical values that can be used by the model; Specific information from the historical data of empty car matching on freight railways is identified as new features.
[0054] Furthermore, the training module includes: The building unit is used to optimize parameters based on sample weights and learning rate to build a basic gradient boosting decision tree model; The iterative unit is used to perform k iterations of training and result verification on the basic gradient boosting decision tree model based on the target training set using the k-fold cross-validation method. The unit is used to identify the best-performing base gradient boosting decision tree model from the validation results of k iterations of training as the empty car matching model for freight railways.
[0055] Furthermore, based on the target training set, the basic gradient boosting decision tree model is trained and validated k times using a k-fold cross-validation method, including: The target training set is randomly divided into k subsets of equal size. In each iteration, one subset is selected as the validation set, and the remaining k-1 subsets are selected as the sub-training set. The basic gradient boosting decision tree model is trained k times based on the sub-training set; The k-th iteration training results of the basic gradient boosting decision tree model are validated based on the validation set corresponding to the sub-training set.
[0056] It should be noted that, for ease of explanation, Figure 2 This example illustrates only the main modules of a freight railway empty car matching system based on the gradient boosting decision tree algorithm. In practical applications, the system may also include modules or components not shown in the figure; the system is not limited to the above module structure, but may also be other module structures that implement the above method embodiments.
[0057] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. As shown in the figure, the electronic device includes a processor and a memory. The processor is used to read and execute programs and instructions stored in the memory, causing the electronic device to perform the above-described method embodiments.
[0058] It should be noted that, for ease of explanation, Figure 3 For illustrative purposes only, the main components of the electronic device are shown. In practical applications, the electronic device may also include components or parts not shown in the figures.
[0059] This application also provides a computer-readable storage medium storing a program or instructions that, when read and executed by a computer, cause the computer to perform the above-described method embodiments.
[0060] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A method for matching empty freight cars on railways based on gradient boosting decision tree algorithm, characterized in that, include: Historical data on empty car matching for freight railways are preprocessed to obtain positive sample data for the empty car matching model of freight railways. The empty car data that were not matched in each daily plan in the positive sample data are concatenated with the daily plan to obtain the negative sample data of the freight railway empty car matching model; The positive and negative sample data of the freight railway empty car matching model are merged to obtain the target training set of the freight railway empty car matching model. Based on the target training set, the gradient boosting decision tree model is trained using the k-fold cross-validation method to obtain the freight railway empty car matching model; The given daily schedule and empty car data are input into the freight railway empty car matching model for matching and prediction, and the allocation results of the daily schedule and empty cars are obtained.
2. The method for matching empty freight cars on railways based on gradient boosting decision tree algorithm according to claim 1, characterized in that, The preprocessing of historical data on empty car matching for freight railways includes: Remove null values from the historical data of empty freight railway car matching. Label and encode the category features carried by the textual information in the historical data of empty freight railway car matching; Convert discrete attributes from historical data on empty car matching for freight railways into numerical values that can be used by the model; Specific information from the historical data of empty car matching on freight railways is identified as new features.
3. The method for matching empty freight trains based on gradient boosting decision tree algorithm according to claim 1, characterized in that, The gradient boosting decision tree model, trained using the k-fold cross-validation method on the target training set, yields a freight railway empty car matching model, including: Based on the optimized parameters of sample weights and learning rate, a basic gradient boosting decision tree model is constructed. Based on the target training set, the k-fold cross-validation method is used to perform k iterations of training and result validation on the basic gradient boosting decision tree model. The basic gradient boosting decision tree model with the best average performance among the validation results of k iterations of training is determined as the empty car matching model for freight railways.
4. The method for matching empty freight cars on railways based on gradient boosting decision tree algorithm according to claim 3, characterized in that, The method of performing k iterations of training and result verification on the basic gradient boosting decision tree model based on the target training set and using k-fold cross-validation includes: The target training set is randomly divided into k subsets of equal size. In each iteration, one subset is selected as the validation set, and the remaining k-1 subsets are selected as the sub-training set. The basic gradient boosting decision tree model is trained k times based on the sub-training set; The k-th iteration training results of the basic gradient boosting decision tree model are validated based on the validation set corresponding to the sub-training set.
5. A freight railway empty car matching system based on gradient boosting decision tree algorithm, characterized in that, include: The processing module is used to preprocess the historical data of empty car matching for freight railways to obtain positive sample data for the empty car matching model of freight railways. The splicing module is used to splice the empty car data that were not matched in each daily plan in the positive sample data with the daily plan to obtain the negative sample data of the freight railway empty car matching model; The merging module is used to merge the positive and negative sample data of the freight railway empty car matching model to obtain the target training set of the freight railway empty car matching model. The training module is used to train the gradient boosting decision tree model based on the target training set using the k-fold cross-validation method, so as to obtain the freight railway empty car matching model. The prediction module is used to input the given daily plan and empty car data into the freight railway empty car matching model for matching and prediction, and obtain the daily plan and empty car allocation results.
6. The freight railway empty car matching system based on gradient boosting decision tree algorithm according to claim 5, characterized in that, The preprocessing of historical data on empty car matching for freight railways includes: Remove null values from the historical data of empty freight railway car matching. Label and encode the category features carried by the textual information in the historical data of empty freight railway car matching; Convert discrete attributes from historical data on empty car matching for freight railways into numerical values that can be used by the model; Specific information from the historical data of empty car matching on freight railways is identified as new features.
7. The freight railway empty car matching system based on gradient boosting decision tree algorithm according to claim 5, characterized in that, The training module includes: The building unit is used to optimize parameters based on sample weights and learning rate to build a basic gradient boosting decision tree model; The iterative unit is used to perform k iterations of training and result verification on the basic gradient boosting decision tree model based on the target training set using the k-fold cross-validation method. The unit is used to identify the best-performing base gradient boosting decision tree model from the validation results of k iterations of training as the empty car matching model for freight railways.
8. The freight railway empty car matching system based on gradient boosting decision tree algorithm according to claim 7, characterized in that, The method of performing k iterations of training and result verification on the basic gradient boosting decision tree model based on the target training set and using k-fold cross-validation includes: The target training set is randomly divided into k subsets of equal size. In each iteration, one subset is selected as the validation set, and the remaining k-1 subsets are selected as the sub-training set. The basic gradient boosting decision tree model is trained k times based on the sub-training set; The k-th iteration training results of the basic gradient boosting decision tree model are validated based on the validation set corresponding to the sub-training set.
9. An electronic device, characterized in that, include: Processor and memory; The processor is coupled to the memory; The processor is configured to read and execute the program or instructions stored in the memory, causing the device to perform the method as described in any one of claims 1-4.
10. A computer-readable storage medium, characterized in that, The device contains a computer program that, when executed by a processor, implements the method as described in any one of claims 1-4.