A data-driven freight train operation prediction method and system
Through the data-driven multi-regression integrated learning model, predicting sudden abnormalities in the freight train transportation network, solving the problem of difficulty in accurately predicting the impact of train transportation networks in the prior art, achieving accurate and timely scheduling and adjustments, and improving the quality of transportation services.
Patent Information
- Application Number
- CN202111069775.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-13
- Publication Date
- 2024-12-10
- Estimated Expiration
- 2041-09-13
AI Technical Summary
It is difficult for the prior art to accurately predict the impact of sudden abnormalities on the transportation network in train transportation networks, resulting in delayed abnormalities being unable to respond and schedule and adjust in a timely manner, affecting the quality of transportation services.
The data-driven freight train operation prediction method is adopted, and through multi-regression integrated learning model training and prediction, combining the time, type and position data of emergencies, the fault impact status of the corresponding turntable is predicted, and the data basis is provided to adjust the train transportation network.
It realizes that when the cause of the abnormality is not clear, it accurately and timely predicts the impact of abnormal conditions on the transportation network, meets the scheduling needs in the transportation production process, avoids the propagation of late abnormalities, and improves the quality of transportation services.
Smart Images
Figure CN113935229B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a rail transportation system, and in particular to a data-driven freight train operation prediction method and system. Background Art
[0002] The operating environment of the railway system is complex. Freight trains in particular have long transportation distances, small train intervals, and high transportation frequencies. During operation, they are inevitably affected by various factors such as the external environment, internal system mechanical failures, and organizational management, and encounter various emergencies, resulting in deviations between the actual arrival time (actual departure time) and the planned arrival time (planned departure time). Severe delays may spread over a large area of the line and local network, affecting the normal execution of the train operation plan and reducing the quality of transportation services.
[0003] In order to respond to sudden anomalies on the train transportation network in a timely manner, respond to sudden anomalies as accurately as possible, and properly arrange other affected vehicles in the train transportation network, it is necessary to evaluate the train operation situation.
[0004] In the prior art, existing train delay studies at home and abroad mostly use traditional mathematical model-driven methods or simulation methods to solve related problems. However, when using these methods to establish relevant models, it is necessary to have an accurate description and understanding of parameters such as the cause of delays, delay distribution, delay propagation mechanism, and delay recovery process. In the actual transportation process, the cause, propagation status, and recovery status of train transportation anomalies are usually unable to obtain accurate conclusions on site. Therefore, the existing assessment of train operation conditions is based on certain assumptions and preconditions. The evaluation models and algorithms provided in the prior art have specific scope of use requirements. Therefore, when the cause of the anomaly is unclear, the existing evaluation model cannot accurately predict the impact of the abnormal condition on the transportation network and cannot meet the scheduling needs in the actual transportation production process. Summary of the invention
[0005] In view of the shortcomings of the prior art, the present invention provides a data-driven freight train operation prediction method and system. The present invention collects and predicts the occurrence time, train number, type, and location of sudden abnormalities, and can timely and accurately know the fault impact of the corresponding turnout, thereby providing a data basis for scheduling and adjusting other vehicles in the train transportation network. The present invention specifically adopts the following technical solutions.
[0006] Firstly, to achieve the above purpose, a data-driven freight train operation prediction method is proposed. The method first completes the training of the multi-regression ensemble learning model through the following first to fourth steps, and then inputs the emergency data actually collected in the freight train transportation network into the trained multi-regression ensemble learning model to obtain the prediction value of the corresponding turnout fault impact condition: the first step is to collect the basic data of various emergencies in the freight train transportation network, sort out the impact status data corresponding to each emergency, and establish a freight train transportation data set; the second step is to clean the data in the freight train transportation data set, remove the abnormal amount and null value in the data set, and The characteristic variables of the data set are extracted, and the characteristic variables corresponding to different types of emergencies are One-Hot encoded and normalized according to their categories to obtain a standard data set; the third step is to divide the standard data set into a training data set and a test data set, and construct a multi-regression ensemble learning model, and use the training data set to train each regression model in the multi-regression ensemble learning model separately until each regression model obtains a prediction value that is consistent with the impact status data of the emergency in the test data set; the fourth step is to set weights for each regression model obtained by training in the third step to obtain a trained multi-regression ensemble learning model.
[0007] Optionally, a data-driven freight train operation prediction method as described in any of the above, wherein, in the first step, the basic data of various emergencies in the freight train transportation network include: the time of occurrence of the emergency, train number, event type, event route, event location, and train speed limit level; the impact status data corresponding to each emergency include: event duration, number of affected intervals, number of affected stations, train delay time, number of delayed trains, and number of suspended trains.
[0008] Optionally, a data-driven freight train operation prediction method as described in any of the above, wherein, in the second step, the feature variables contain data corresponding to: event type, event route, train speed limit level and number of trains; in the second step, the feature variables corresponding to different types of emergencies obtained by extracting the feature variables are One-Hot encoded and normalized according to their categories, and the specific steps of obtaining the standard data set include: One-Hot encoding the data corresponding to the event type, train number, event route and event location; numerical mapping of the data at the time of occurrence of the emergency; and normalization of the data corresponding to the train speed limit level and the number of trains according to the range of their maximum and minimum values.
[0009] Optionally, in any of the data-driven freight train operation prediction methods described above, wherein, in the third step, each regression model in the multi-regression ensemble learning model includes at least: a ridge regression model, a lasso regression model, and an elastic regression model.
[0010] Optionally, in any of the above data-driven freight train operation prediction methods, the ridge regression model uses minimizing the residual sum of squares as the objective Among them, θ represents the regularization parameter, β represents the regression parameter, Y represents the actual value of the freight train running time, X represents the feature variable, and P represents the number of samples.
[0011] Optionally, in any of the above data-driven freight train operation prediction methods, the Lasso regression model uses an absolute value as a penalty term in the objective function, and the penalty term Among them, θ represents the regularization parameter, β represents the regression parameter, Y represents the actual value of the freight train running time, X represents the feature variable, and P represents the number of samples.
[0012] Optionally, in any of the above data-driven freight train operation prediction methods, the elastic regression model controls the regularization parameter in the form of a mixing ratio by compromise in the objective function, and the objective function Among them, α represents the proportion of the first type of penalty function, β represents the regression parameter, Y represents the actual value of the freight train running time, X represents the characteristic variable, and P represents the number of samples.
[0013] At the same time, in order to achieve the above-mentioned purpose, the present invention also provides a data-driven freight train operation prediction system, which includes: a data acquisition unit, used to collect basic data of various emergencies in the freight train transportation network, and obtain impact status data corresponding to each emergency event; a data set preprocessing unit, used to establish a freight train transportation data set according to the basic data and impact status data collected by the data acquisition unit; a feature variable extraction unit, used to clean up each data in the freight train transportation data set, eliminate abnormal quantities and null values in the data set, extract feature variables from the cleaned data set, and perform One-Hash on the feature variables corresponding to different types of emergencies according to their categories. ot encoding and normalization processing to obtain a standard data set; a multi-regression ensemble learning model, which uses a portion of data separated from the standard data set as a training data set to train each regression model in the multi-regression ensemble learning model separately, until each regression model obtains a prediction value that is consistent with the impact status of the emergency event corresponding to the remaining data in the standard data set, and then sets weights for each regression model obtained by training to obtain a trained multi-regression ensemble learning model; the multi-regression ensemble learning model is also used to receive emergency event data actually collected in the freight train transportation network, and calculates the prediction value of the fault impact status of the corresponding turnout according to the trained multi-regression ensemble learning model.
[0014] Optionally, in a data-driven freight train operation prediction system as described above, the data acquisition unit is provided with a data interface for receiving the following types of data: the time of occurrence of the emergency, the train number, the type of event, the line where the event occurred, the location where the event occurred, and the train speed limit level.
[0015] Optionally, in a data-driven freight train operation prediction system as described in any of the above, the characteristic variable extraction unit further includes: a ne-Hot encoding module for One-Hot encoding the data corresponding to the event type, train number, event line, and event location; a numerical mapping module for numerically mapping the data at the time of occurrence of the emergency; and a normalization processing module for normalizing the data corresponding to the train speed limit level and the number of trains according to the range of their maximum and minimum values.
[0016] Beneficial Effects
[0017] The present invention trains the ridge regression model, the Lasso regression model, and the elastic regression model respectively, and then sets corresponding weights for each model respectively, and uses the weighted regression models to construct a multi-regression integrated learning model that can make up for the output deviation of a single model and can directly fit various emergencies in the freight train transportation network by setting weight coefficients. Therefore, the predicted value of the fault impact condition of the corresponding turnout obtained by the present application through the calculation of the model can ensure the accuracy and timeliness of the model output when the cause and mechanism of the emergency are unclear. The present application can more accurately predict the impact of abnormal conditions on the transportation network and meet the scheduling needs in the transportation production process.
[0018] Further, in order to cooperate with the above-mentioned model to comprehensively respond to the basic data of various emergencies in the freight train transportation network, and facilitate the dispatching center to perform corresponding dispatching according to the predicted value of the fault impact status of the corresponding turnout output by the model, the present application performs One-Hot encoding on the data corresponding to the event type, train number, event occurrence line, and event occurrence location, and numerically maps the data at the time of the emergency, and normalizes the data corresponding to the train speed limit level and the number of trains according to the range of their maximum and minimum values to obtain the characteristic variables that can comprehensively characterize the data of the event type, event occurrence line, train speed limit level, and number of trains. Thus, the present application can accurately obtain the impact status data of the event duration, the number of affected intervals, the number of affected stations, the train delay time, the number of delayed trains, and the number of suspended trains corresponding to the emergency by training the model and combining the above-mentioned characteristic variables. Thus, the dispatching center can directly adjust the operation of other trains in the freight train transportation network in a timely manner according to the above-mentioned data, and avoid the delay anomaly caused by the emergency from spreading in the line and local network over a large area, affecting the normal execution of the train operation plan and reducing the quality of transportation services.
[0019] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:
[0021] Figure 1 It is a flow chart of the data-driven freight train operation prediction method of the present invention;
[0022] Figure 2 It is a flow chart of the steps of training the multi-regression ensemble learning model in the present invention. DETAILED DESCRIPTION
[0023] In order to make the purpose and technical solution of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment 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.
[0024] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.
[0025] Figure 1 A data-driven freight train operation prediction method according to the present invention is firstly Figure 2 The process steps shown in the figure complete the construction and training of the multi-regression integrated learning model based on the basic data of various emergencies in the freight train transportation network, and then obtain the predicted value of the fault impact condition of the corresponding turnout by inputting the emergency data actually collected in the freight train transportation network into the trained multi-regression integrated learning model.
[0026] Specific reference Figure 2 , the training of the above multi-regression ensemble learning model specifically includes the following steps:
[0027] The first step is to collect basic data of various emergencies in the freight train transportation network, such as the time of occurrence of the emergency, train number, event type, event route, event location, and train speed limit level. At the same time, the impact status data corresponding to each emergency, such as event duration, number of affected sections, number of affected stations, train delay time, number of delayed trains, number of suspended trains, etc., are sorted out, and the basic data and impact status data are stored one by one to establish a freight train transportation data set;
[0028] The second step is to clean the data in the freight train transportation data set, remove the anomalies and null values in the data set, and extract the characteristic variables of the cleaned data set. In order to accurately characterize various emergencies in the freight train transportation network, the characteristic variables contain data corresponding to: event type, event line, train speed limit level, and number of trains. Then, the extracted characteristic variables corresponding to different types of emergencies are One-Hot encoded and normalized according to their categories to obtain a standard data set;
[0029] The third step is to divide the standard data set into a training data set and a test data set, construct a multi-regression ensemble learning model including a ridge regression model, a lasso regression model, and an elastic regression model, and use the training data set to train each regression model in the multi-regression ensemble learning model separately until each regression model obtains a prediction value that is consistent with the impact status data of the emergency event in the test data set, so as to complete the independent training of each regression model and obtain three groups of models that can respectively characterize the impact status of various emergencies in different freight train transportation networks;
[0030] The fourth step is to set the weights of the three groups of regression models obtained by training in the third step. According to the probability of the occurrence of the emergency, the model that can more accurately predict the emergency with a higher probability can be set to have a larger weight, and the model that can more accurately predict the emergency with a smaller probability can be set to have a smaller weight; or, according to the accuracy of the model, the weight can be set according to the characteristic variables corresponding to the emergency matching the model; or according to other prior laws or experience values, the weights of each regression model can be set accordingly, and the multi-regression ensemble learning model with the corresponding weights can be output. The multi-regression ensemble learning model outputted in this way can accurately match different types of emergencies to accurately evaluate them.
[0031] In some implementations, in the second step, the characteristic variables corresponding to different types of emergencies obtained by extracting the characteristic variables are One-Hot encoded and normalized according to their categories. The specific steps for obtaining the standard data set can be set accordingly according to the different meanings of the characteristic variable data:
[0032] One-hot encoding is performed on the data corresponding to the event type, train number, event route, and event location;
[0033] Numerical mapping of data at the time of occurrence of emergencies;
[0034] The data corresponding to the train speed limit level and the number of trains are normalized according to the range of their maximum and minimum values.
[0035] The standard data set thus established can achieve data sharing among various types of regression models, and will not affect the calculation accuracy of each model due to conversion deviations in the input data of each model.
[0036] Among them, the ridge regression model used in the above multi-regression ensemble learning model can be specifically selected by minimizing the residual sum of squares as the target Among them, θ represents the regularization parameter, β represents the regression parameter, Y represents the actual value of the freight train running time, X represents the feature variable, and P represents the number of samples.
[0037] The Lasso regression model used in the above multi-regression ensemble learning model can be specifically selected by using the absolute value as the penalty term in the objective function. Among them, θ represents the regularization parameter, β represents the regression parameter, Y represents the actual value of the freight train running time, X represents the feature variable, and P represents the number of samples.
[0038] The elastic regression model used in the above multi-regression ensemble learning model can specifically choose to control the regularization parameter in the form of a mixed ratio in the objective function through a compromise method. Its objective function Among them, α represents the proportion of the first type of penalty function, β represents the regression parameter, Y represents the actual value of the freight train running time, X represents the characteristic variable, and P represents the number of samples.
[0039] Therefore, the present invention can calculate the predicted values of the corresponding various fault impact conditions of the turnout through the model, so that when the cause of the emergency and the mechanism of the emergency are not clear, the prior model can be trained to ensure that it can accurately and timely output the prediction results. According to the prediction results of the model, the present application can more accurately evaluate the impact of abnormal conditions caused by emergencies on the transportation network, and make timely scheduling adjustments to ensure that transportation production is stable and orderly.
[0040] The above is only an embodiment of the present invention, and its description is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.
Claims
1. A data-driven freight train operation prediction method, characterized in that: First, complete the training of the multi-regression ensemble learning model through the following steps 1 to 4, and then input the emergency data actually collected in the freight train transportation network into the trained multi-regression ensemble learning model to obtain the predicted value of the corresponding turnout fault impact condition: The first step is to collect basic data of various emergencies in the freight train transportation network, sort out the impact status data corresponding to each emergency, and establish a freight train transportation data set; The second step is to clean the data in the freight train transportation data set, remove the outliers and null values in the data set, extract feature variables from the cleaned data set, and perform One-Hot encoding and normalization on the feature variables corresponding to different types of emergencies according to their categories to obtain a standard data set. The third step is to divide the standard data set into a training data set and a test data set, construct a multi-regression ensemble learning model, and use the training data set to train each regression model in the multi-regression ensemble learning model separately until each regression model obtains a prediction value that is consistent with the impact status data of the emergency event in the test data set; The fourth step is to set the weights of the regression models obtained through training in the third step to obtain a trained multi-regression ensemble learning model.
2. The data-driven freight train operation prediction method according to claim 1, characterized in that: In the first step, the basic data of various emergencies in the freight train transportation network include: the time of occurrence of the emergency, the train number, the event type, the event line, the event location, and the train speed limit level; The impact status data corresponding to each emergency event include: event duration, number of affected sections, number of affected stations, train delay time, number of delayed trains, and number of suspended trains.
3. The data-driven freight train operation prediction method according to claims 1-2, characterized in that: In the second step, the characteristic variables include data corresponding to: event type, event occurrence line, train speed limit level and train number; In the second step, the characteristic variables corresponding to different types of emergencies obtained by feature variable extraction are One-Hot encoded and normalized according to their categories. The specific steps to obtain the standard data set include: One-hot encoding is performed on the data corresponding to the event type, train number, event route, and event location; Numerical mapping of data at the time of occurrence of emergencies; The data corresponding to the train speed limit level and the number of trains are normalized according to the range of their maximum and minimum values.
4. The data-driven freight train operation prediction method according to claims 1-3, characterized in that: In the third step, each regression model in the multi-regression ensemble learning model includes at least: a ridge regression model, a lasso regression model, and an elastic regression model.
5. The data-driven freight train operation prediction method according to claim 4, characterized in that: The ridge regression model uses minimizing the residual sum of squares as its goal Among them, θ represents the regularization parameter, β represents the regression parameter, Y represents the actual value of the freight train running time, X represents the feature variable, and P represents the number of samples.
6. The data-driven freight train operation prediction method according to claim 4, characterized in that: The Lasso regression model uses the absolute value as the penalty term in the objective function. Among them, θ represents the regularization parameter, β represents the regression parameter, Y represents the actual value of the freight train running time, X represents the feature variable, and P represents the number of samples.
7. The data-driven freight train operation prediction method according to claim 4, characterized in that: The elastic regression model uses a compromise in the objective function to control the regularization parameter in the form of a mixing ratio. Among them, α represents the proportion of the first type of penalty function, β represents the regression parameter, Y represents the actual value of the freight train running time, X represents the characteristic variable, and P represents the number of samples.
8. A data-driven freight train operation prediction system, characterized in that: include: A data collection unit is used to collect basic data of various emergencies in the freight train transportation network and obtain impact status data corresponding to each emergency event; A data set preprocessing unit, used to establish a freight train transportation data set based on the basic data and influencing status data collected by the data collection unit; The characteristic variable extraction unit is used to clean the data in the freight train transportation data set, remove the abnormalities and null values in the data set, extract characteristic variables from the cleaned data set, and perform One-Hot encoding and normalization processing on the characteristic variables corresponding to different types of emergencies according to their categories to obtain a standard data set; A multi-regression ensemble learning model, which uses a portion of data separated from a standard data set as a training data set to train each regression model in the multi-regression ensemble learning model separately, until each regression model obtains a prediction value that is consistent with the impact status of the emergency event corresponding to the remaining data in the standard data set, and then sets weights for each regression model obtained by training to obtain a trained multi-regression ensemble learning model; The multi-regression integrated learning model is also used to receive emergency event data actually collected in the freight train transportation network, and calculate the predicted value of the fault impact status of the corresponding turnout based on the trained multi-regression integrated learning model.
9. The data-driven freight train operation prediction system according to claim 8, characterized in that: The data acquisition unit is provided with a data interface for receiving the following types of data: the time of occurrence of the emergency, the train number, the event type, the event line, the event location, and the train speed limit level.
10. The data-driven freight train operation prediction system according to claims 8-9, characterized in that: The characteristic variable extraction unit further includes: One-Hot encoding module, used to perform One-Hot encoding on the data corresponding to the event type, train number, event line, and event location; A numerical mapping module is used to perform numerical mapping on the data at the time of occurrence of the emergency; The normalization processing module is used to normalize the data corresponding to the train speed limit level and the number of trains according to the range of their maximum and minimum values.