High-speed railway overhead line system dropper arrangement optimization method based on machine learning technology

Through machine learning technology, a nonlinear mapping model of the position and contact pressure of the hanging string is established, combined with particle swarm optimization algorithm, the hanging string layout of the high-speed railway contact network is optimized, which solves the problem of the hanging string withstanding impact and vibration during high-speed train operation, and extends the The lifespan of the strings and improves contact quality.

CN118917210APending Publication Date: 2024-11-08CHINA STATE RAILWAY GRP CO LTD +3
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411210934.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The high-speed railway contact network hanging strings suffer strong impact and vibration during the operation of high-speed trains, resulting in fatigue damage to the metal hanging strings. The existing technology is difficult to effectively optimize the hanging string layout to extend its life, and the project volume is large and the cost is high.

Method used

Using a machine learning-based method, a nonlinear mapping model of the position parameters of the hanging string and the contact pressure is established, combined with neural network and particle swarm optimization algorithm, the hanging string layout is optimized to reduce contact pressure fluctuations and extend the service period of the hanging string.

Benefits of technology

By optimizing the layout of the hanging string, the vibration and elastic unevenness of the contact network are reduced, the contact quality is improved, the fatigue life of the hanging string is extended, and a more reasonable contact network structure design is achieved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917210A_ABST
    Figure CN118917210A_ABST
Patent Text Reader

Abstract

The invention provides a machine learning technology-based optimization method for dropper arrangement of a high-speed railway overhead line system, which is realized by combining a neural network and an optimization algorithm and further quickly and reasonably adjusting dropper layout parameters. When the structure of the overhead line system is optimized, the position of the dropper in a span is used as an input quantity, the contact pressure standard deviation is used as a target output, and the contact pressure fluctuation is reduced as much as possible by adjusting the position of the dropper in the span. The objective of the invention is to provide a reasonable optimization method for the layout design of the catenary dropper and improve the fatigue life of the dropper in the service period. A mapping model is established through correlation between catenary dropper position information and contact pressure in combination with a machine learning means, an optimization result is searched, and then the optimization result is simulated and verified.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of rail transit technology, and in particular to a method for optimizing the arrangement of high-speed railway contact network hanger strings based on machine learning technology. Background Art

[0002] The overhead contact system is an important part of high-speed railways. High-speed trains rely on the sliding friction between the roof pantograph and the overhead contact wire to obtain electrical energy. When the overhead contact is stationary, the suspension string only bears the tensile stress caused by the gravity of the contact wire. However, when the train passes by the suspension string, the upward lifting of the contact wire by the high-speed sliding pantograph will inevitably cause the suspension string to bear a strong instantaneous impact. After the train passes, the continuous vibration of the overhead contact system will cause the suspension string to repeatedly bear alternating stress. Fatigue damage is inevitable when the metal suspension string is in such a service state for a long time.

[0003] Different layout schemes of the suspension strings within the catenary span will affect the dynamic interaction results of the bow and catenary. It is possible to consider improving the stress condition of the suspension strings during service by adjusting the layout of the suspension strings, in order to extend the fatigue life of the suspension strings. However, the catenary span is long, and there are many suspension strings in one span, so the number of layout schemes will be extremely large. In previous technologies, the layout schemes of the suspension strings were mostly adjusted based on engineering experience or multiple tests. This method has a huge amount of engineering work, and the cost of trying out all the schemes and then selecting the most reasonable installation method is unacceptable. This method considers combining machine learning to establish a nonlinear mapping model of the suspension string position parameters and contact pressure, and optimizes the layout of the suspension strings in the catenary in combination with engineering needs. Summary of the invention

[0004] The purpose of the present invention is to provide a reasonable optimization method for the layout design of overhead contact wires, so as to improve the fatigue life of the wires during their service life. Through the correlation between the location information of overhead contact wires and the contact pressure, a mapping model is established by combining machine learning methods, and the optimization effect is verified by simulation after searching for the optimization results.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for optimizing the arrangement of high-speed railway contact network droppers based on machine learning technology comprises the following steps:

[0007] First, for the current or designed pantograph model, according to the number of suspension strings within the catenary span;

[0008] Determine the dimensions of the neural network input layer and the boundaries of each input parameter;

[0009] The samples of the neural network training set are formulated according to the optimization target; the time series of the pantograph-catenary contact pressure when the train is running under the corresponding working conditions is obtained according to the formulated training samples, and the standard deviation of the target series is calculated to determine the value of the neural network output parameter; by performing the above steps within the boundary of each parameter of the input layer, a certain number of test sets are randomly collected to verify the established nonlinear mapping model between the catenary suspension string arrangement and the contact force statistical value;

[0010] Then, the training and prediction samples collected above are used to continuously adjust the hyperparameters in the model to balance the model's fitting effect and generalization performance, and to establish a neural network proxy model;

[0011] Finally, the intelligent optimization algorithm is used to find the optimal solution for the neural network model mapping.

[0012] The specific steps are:

[0013] Step 1: For the initial design or the existing line contact network structure, obtain the time series of the pantograph-catenary contact pressure when the train is running under the parameters, and calculate the standard deviation of the contact pressure.

[0014] The calculation formula of contact force standard deviation is as follows:

[0015]

[0016] In formula (1), σ represents the standard deviation of the sequential contact pressure, F i It represents the contact pressure value (N) when the pantograph runs to the contact network sampling point, where n is the number of contact network sampling points.

[0017] The minimum standard deviation is taken as the optimization goal.

[0018] Step 2: Analyze the initial contact network structure and determine the dimension of the proxy model input layer and the upper and lower boundaries of the input parameters in combination with the restrictions on the contact network dropper design in the project.

[0019] Step 3: Prepare samples of the neural network training set according to the optimization target, obtain the time series of the pantograph-catenary contact pressure when the train is running under the corresponding working conditions according to the prepared training samples, and calculate the standard deviation of the target sequence. At the same time, randomly collect a certain number of test sets to verify the established nonlinear mapping model between the catenary suspension string arrangement and the contact force statistics;

[0020] Step 4: Use the training and prediction samples collected above to continuously adjust the hyperparameters in the model to balance the model's fitting effect and generalization performance, such as the number of hidden layers and nodes, the required training accuracy, the maximum number of training times, the learning rate, etc., to establish a neural network proxy model; use the mean absolute percentage error MAPE as an indicator to evaluate the accuracy of the proxy model;

[0021]

[0022] Where n is the number of test points, y i is the actual value, is the predicted value; the closer MAPE is to 0, the smaller the error between the predicted value and the true value, and the more accurate the prediction result.

[0023] Step 5: Combine the particle swarm optimization algorithm to find the optimal solution for the neural network model mapping. First, initialize the population number, inertia weight, learning rate, and number of iterations in the algorithm, randomly set the position and speed of each particle, calculate the fitness value according to the fitness function, record and retain the historical optimal value and the global optimal value, and determine whether the maximum number of iterations has been reached. If not, continuously update the particle swarm position and speed until the preset value is reached to obtain the optimal input parameters;

[0024] Step 6: Use the optimal input parameters obtained as the input of the simulation model, calculate and compare the static characteristics and dynamic response of the model after the optimized layout. If the model calculation results do not meet the following engineering requirements, the optimization is considered to have failed:

[0025] (1) There should be no contact loss between the bow and the catenary;

[0026] (2) The contact force between the bow and the catenary is within the standard limit;

[0027] (3) The lifting amount of the positioning point is greater than 120 mm;

[0028] (4) The penetration depth of the pantograph is greater than 80 mm.

[0029] The present invention provides a reasonable optimization method for the layout design of high-speed railway overhead contact wire, which is combined with neural network and optimization algorithm, and then realized by quickly and reasonably adjusting the layout parameters of the wire. When optimizing the overhead contact wire structure, the position of the wire in a span is used as input, and the standard deviation of the contact pressure is used as the target output. The contact pressure fluctuation is reduced as much as possible by adjusting the position of the wire in the span. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A diagram of the contact network structure parameters of an embodiment;

[0031] Figure 2 A diagram showing the structural parameters of a pantograph in an embodiment;

[0032] Figure 3 This is a diagram of prediction results using machine learning in an embodiment;

[0033] Figure 4 This is a comparison diagram of the contact network shapes before and after optimization of the embodiment;

[0034] Figure 5 This is a stiffness comparison diagram before and after optimization of the embodiment;

[0035] Figure 6 This is a comparison diagram of the bow-net contact force before and after optimization of the embodiment. DETAILED DESCRIPTION

[0036] The technical solution of the present invention will be described in detail below in conjunction with the accompanying drawings in the examples of the present invention.

[0037] The optimization method of high-speed railway contact network dropper arrangement based on machine learning technology includes the following steps:

[0038] Step 1: For the initial design or the existing line contact network structure, obtain the time series of the pantograph-catenary contact pressure when the train is running under the parameters, and calculate the standard deviation of the contact pressure. Figure 1 , Figure 2 The pantograph-catenary structural parameters shown are used as an example of this embodiment, that is, an initial design scheme.

[0039] The calculation formula of contact force standard deviation is as follows:

[0040]

[0041] In formula (1), σ represents the standard deviation of the sequential contact pressure, F i It represents the contact pressure value (N) when the pantograph runs to the contact network sampling point, and n is the number of contact network sampling points. During the operation of the train, the continuous vibration between the pantograph and the contact network causes the high-frequency vibration of the suspension string, which accelerates the fatigue damage of the suspension string. By reducing the vibration between the pantograph and the contact network, the damage to the suspension string during the operation of the train can be effectively reduced, and the vibration between the pantograph and the contact network can be accurately represented by the standard deviation of the contact force between the pantograph and the contact network. The larger the value, the more severe the vibration between the pantograph and the contact network, and the smaller the value, the milder the vibration between the pantograph and the contact network. Therefore, the minimum standard deviation is taken as the optimization target of this method.

[0042] Step 2: Analyze the initial contact network structure and determine the dimensions of the proxy model input layer and the upper and lower boundaries of the input parameters in combination with the constraints on the contact network dropper design in the project. For example, in this embodiment, there are six droppers in the span of the structure diagram. Considering the symmetrical distribution and design requirements of the contact network structure, the position information x of the first three droppers in the span is i (i=1,2,3) is used as the input of the proxy model, and the ranges are 2m≤x1≤5m, 5m≤x2≤15m, 15m≤x3<25m respectively. The output of the proxy model is the contact force standard deviation σ.

[0043] Step 3: Prepare samples of the neural network training set according to the optimization target, obtain the time series of the pantograph-catenary contact pressure when the train is running under the corresponding working conditions according to the prepared training samples, and calculate the standard deviation of the target sequence. At the same time, randomly collect a certain number of test sets to verify the established nonlinear mapping model between the catenary suspension string arrangement and the contact force statistics;

[0044] Step 4: Use the training and prediction samples collected above to continuously adjust the hyperparameters in the model to balance the model's fitting effect and generalization performance, such as the number of hidden layers and nodes, training accuracy requirements, maximum number of training times, learning rate, etc., to establish a neural network proxy model. The final prediction effect is as follows: Figure 3 As shown in the figure, the mean absolute percentage error MAPE is used as an indicator to evaluate the accuracy of the proxy model;

[0045]

[0046] Where n is the number of test points, y i is the actual value, is the predicted value; the closer MAPE is to 0, the smaller the error between the predicted value and the true value, and the more accurate the prediction result.

[0047] Step 5: Combine the particle swarm optimization algorithm to find the optimal solution for the neural network model mapping. First, initialize the population number, inertia weight, learning rate, and number of iterations in the algorithm, randomly set the position and speed of each particle, calculate the fitness value according to the fitness function, record and retain the historical optimal value and the global optimal value, and determine whether the maximum number of iterations has been reached. If not, continuously update the particle swarm position and speed until the preset value is reached to obtain the optimal input parameters;

[0048] Step 6: Use the optimal input parameters obtained as the input of the simulation model, calculate and compare the static characteristics and dynamic response of the model after the optimized layout. If the model calculation results do not meet the following engineering requirements, the optimization is considered to have failed:

[0049] (1) There should be no contact loss between the bow and the catenary;

[0050] (2) The contact force between the bow and the catenary is within the standard limit;

[0051] (3) The lifting amount of the positioning point is greater than 120 mm;

[0052] (4) The depth of pantograph penetration is greater than 80 mm;

[0053] Figure 4 , Figure 5 , Figure 6By comparing the results of the bow and net before and after optimization in the embodiment, it can be found that after optimizing the suspension string layout, this method can reduce the elastic unevenness of the contact network and improve the contact quality of the bow and net, improve the stress condition of the suspension string during service, and is expected to extend the fatigue life of the suspension string.

Claims

1. A method for optimizing the arrangement of high-speed railway overhead contact wires based on machine learning technology, characterized in that: The following steps are involved: First, for the current or designed pantograph model, according to the number of suspension strings within the catenary span; Determine the dimensions of the neural network input layer and the boundaries of each input parameter; The samples of the neural network training set are formulated according to the optimization target; the time series of the pantograph-catenary contact pressure when the train is running under the corresponding working conditions is obtained according to the formulated training samples, and the standard deviation of the target series is calculated to determine the value of the neural network output parameter; by performing the above steps within the boundary of each parameter of the input layer, a certain number of test sets are randomly collected to verify the established nonlinear mapping model between the catenary suspension string arrangement and the contact force statistical value; Then, the training and prediction samples collected above are used to continuously adjust the hyperparameters in the model to balance the model's fitting effect and generalization performance, and to establish a neural network proxy model; Finally, the intelligent optimization algorithm is used to find the optimal solution for the neural network model mapping.

2. The method for optimizing the arrangement of high-speed railway overhead contact network droppers based on machine learning technology according to claim 1, characterized in that: The following steps are involved: Step 1: For the initial design scheme or the existing line contact network structure, obtain the time series of the pantograph-catenary contact pressure when the train is running under the parameters, and calculate the standard deviation of the contact pressure; The minimum standard deviation is taken as the optimization goal; Step 2: Analyze the initial contact network structure and determine the dimension of the proxy model input layer and the upper and lower boundaries of the input parameters in combination with the constraints on the contact network dropper design in the project; Step 3: Prepare samples of the neural network training set according to the optimization target, obtain the time series of the pantograph-catenary contact pressure when the train is running under the corresponding working conditions according to the prepared training samples, and calculate the standard deviation of the target sequence. At the same time, randomly collect a certain number of test sets to verify the established nonlinear mapping model between the catenary suspension string arrangement and the contact force statistics; Step 4: Use the training and prediction samples collected above to continuously adjust the hyperparameters in the model to balance the model's fitting effect and generalization performance, and establish a neural network proxy model; use the mean absolute percentage error (MAPE) as an indicator to evaluate the accuracy of the proxy model; Step 5: Combine the particle swarm optimization algorithm to find the optimal solution for the neural network model mapping. First, initialize the population number, inertia weight, learning rate, and number of iterations in the algorithm, randomly set the position and speed of each particle, calculate the fitness value according to the fitness function, record and retain the historical optimal value and the global optimal value, and determine whether the maximum number of iterations has been reached. If not, continuously update the particle swarm position and speed until the preset value is reached to obtain the optimal input parameters; Step 6: Use the optimal input parameters obtained as the input of the simulation model, calculate and compare the static characteristics and dynamic response of the model after the optimized layout. If the model calculation results do not meet the engineering requirements, the optimization is considered to have failed.

3. The method for optimizing the arrangement of high-speed railway overhead contact network droppers based on machine learning technology according to claim 2 is characterized in that: In step 1, the formula for calculating the standard deviation of contact force is as follows: In formula (1), σ represents the standard deviation of the sequential contact pressure, F i It represents the contact pressure value (N) when the pantograph runs to the contact network sampling point, where n is the number of contact network sampling points.

4. The method for optimizing the arrangement of high-speed railway overhead contact network droppers based on machine learning technology according to claim 2, characterized in that: In step 4, the mean absolute percentage error MAPE is; Where n is the number of test points, y i is the actual value, is the predicted value; the closer MAPE is to 0, the smaller the error between the predicted value and the true value, and the more accurate the prediction result.

5. The method for optimizing the arrangement of high-speed railway overhead contact network droppers based on machine learning technology according to claim 2, characterized in that: In step 6, engineering requirements include: (1) There should be no contact loss between the bow and the catenary; (2) The contact force between the bow and the catenary is within the standard limit; (3) The lifting amount of the positioning point is greater than 120 mm; (4) The penetration depth of the pantograph is greater than 80 mm.

Citation Information

Cited By

  • Pantograph control system of railway vehicle

    CN119376325A

  • Method and system for optimizing dropper arrangement of high-speed railway overhead line system based on machine learning

    CN119989951A

  • Railway four-electricity intelligent construction monitoring system based on multi-source data fusion

    CN121386573A