Steel rail impedance spectrum construction method based on ensemble learning
Through machine learning regression analysis algorithm and integrated learning method, the rail impedance spectrum is constructed, which solves the problem of inaccurate simulation of rail impedance characteristics in the existing technology, and realizes efficient and accurate construction of rail impedance spectrum, reducing detection costs.
Patent Information
- Application Number
- CN202411627042.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-14
- Publication Date
- 2025-03-07
AI Technical Summary
The prior art is difficult to accurately simulate the impedance characteristics of rails in different scenarios and frequencies, resulting in high functional detection cost and impracticality of rail break monitoring equipment.
The machine learning regression analysis algorithm is adopted, combined with the idea of integrated learning, and the random forest and gradient enhancement regression tree algorithm are used to build an equivalent calculation model for rail impedance characteristics and establish a rail impedance spectrum.
Through a small amount of on-site measured data, the continuous rail impedance spectrum is fitted, which reduces the uncertainty and testing costs of on-site measured, and improves the detection accuracy and efficiency of the rail-break monitoring equipment.
Smart Images
Figure CN119577692A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of railway transportation safety assurance and relates to a method for constructing a rail impedance spectrum based on ensemble learning. Background Art
[0002] Rail breakage or electrical disconnection is a major safety hazard in railway transportation, which can lead to serious accidents such as train derailment and overturning. Therefore, rail break monitoring technology that can continuously monitor the status of rails in real time is a necessary measure to ensure driving safety.
[0003] Electrical circuit type broken rail monitoring equipment is widely used due to its simple structure, low cost, easy use, low power consumption, and low false alarm and missed alarm rates. The equipment uses the rail as part of the circuit and judges the integrity of the rail by analyzing the amplitude, frequency, phase and other information of the electrical signal in the rail. Before the broken rail monitoring equipment is put into use, it is necessary to test its functions and performance. However, on-site testing requires a lot of manpower and time costs, and may affect the operation of existing lines, which is unrealistic. Laboratory testing methods are usually used, using simulation disks to simulate actual rails. Since the rail is an inductive element, its impedance characteristics are closely related to the frequency of the electrical signal.
[0004] Therefore, it is crucial to study the impedance transmission characteristic parameters of rails at different frequencies and construct a rail impedance model in order to detect and verify the functions of electrical circuit-type broken rail monitoring equipment. Summary of the invention
[0005] In order to more accurately simulate the rail impedance characteristics, the present invention measures the rail impedance data in different scenarios and at different frequencies. The present invention uses a machine learning regression analysis algorithm to establish an equivalent calculation model for the rail impedance characteristics.
[0006] The technical solution of the present invention is a method for constructing a rail impedance spectrum based on ensemble learning. The key is that the rail impedance is composed of two parts: resistance and inductance. The specific steps of the construction method include:
[0007] S1. Construct and preprocess the dataset:
[0008] Obtain rail resistance and inductance data sets, preprocess the data sets, and divide the preprocessed data sets into a training set and a test set;
[0009] S2. Construct resistance calculation model and inductance calculation model:
[0010] The rail resistance and inductance data sets are iteratively trained using the random forest algorithm and the gradient boosting regression tree algorithm to obtain a resistance calculation model that associates resistance with frequency and an inductance calculation model that associates inductance with frequency.
[0011] S3. The required continuous frequency range is used as the input of the resistance calculation model and the inductance calculation model, and a continuous rail impedance spectrum is obtained through calculation.
[0012] Furthermore, in the above step S2, hyperparameters are set before building the resistance calculation model and the inductance calculation model. The hyperparameters include hyperparameters of the random forest algorithm and hyperparameters of the gradient boosting regression tree algorithm:
[0013] The hyperparameters of the random forest algorithm mentioned above include the number of weak learners, the maximum depth, and the maximum number of features;
[0014] The hyperparameters of the above gradient boosted regression tree algorithm include learning rate, number of weak learners and maximum depth.
[0015] Furthermore, the above step S2 also includes:
[0016] The hyperparameter space composed of the hyperparameters is selected and optimized by using the training set to obtain a rail impedance calculation model;
[0017] The above selection is to pre-set the value range of the hyperparameter, and the optimization is achieved through grid search and cross validation;
[0018] The test set is input into the resistance calculation model and the inductance calculation model, and the performance of the rail impedance calculation model is evaluated according to the calculation output results to determine whether the evaluation results meet the requirements. If not, S2 is executed again; if it meets the requirements, S3 is executed.
[0019] Specifically, the evaluation indicators of the above performance evaluation include a root mean square error and a determination coefficient. The above compliance requirement is that the root mean square error value is lower than 0.02 and the determination coefficient value is higher than 0.98.
[0020] More specifically, the above mean square error and determination coefficient calculation formulas are as follows:
[0021]
[0022]
[0023] In formula 1 and formula 2, MSE represents mean square error, R 2 represents the coefficient of determination, n represents the number of samples in the test set, i represents the sequence number of the test set data, and y i Indicates the measured value, represents the model output value, Represents the average of the measured values of the test set.
[0024] Furthermore, in the above step S1, the characteristics of the above data set include signal frequency, rail resistance and rail inductance; and the above data set is obtained by field measurement.
[0025] Specifically, the above-mentioned field measurement adopts the double short-circuit method to measure the rail parameters at the railway application site, and the measured rail section does not contain compensation capacitors, rail lead wires and other equipment.
[0026] Furthermore, in the above step S1, the above preprocessing is to perform outlier processing on the data in the data set.
[0027] Specifically, the above-mentioned method for handling outliers is based on the 3σ criterion or the 6σ criterion, and data points that exceed the criterion judgment range are judged as outliers.
[0028] The beneficial effects of the present invention are:
[0029] The present invention utilizes a machine learning regression analysis algorithm to establish an equivalent calculation model for rail impedance characteristics, and obtains an impedance calculation model with frequency universality and a continuous impedance spectrum of the rail through partial discrete impedance data. With the help of this model, the application conditions of the broken rail monitoring equipment are used as input to obtain the impedance parameters of the signal when it is transmitted on the rail.
[0030] The rail impedance calculation model and impedance spectrum calculation method of the present invention can fit and obtain a continuous rail impedance spectrum with only a small amount of field measured data, thereby reducing the uncertainty of field measurement and the test cost.
[0031] The ensemble learning method proposed in the present invention is based on two ensemble ideas, bagging and boosting, and specifically includes two algorithms (random forest and gradient boosting decision tree). It can compare the advantages and disadvantages of prediction models and select the optimal model for rail impedance parameter calculation, thus avoiding the problem of insufficient performance of a single algorithm (such as BP neural network model). BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 It is the rail continuous impedance spectrum-rail resistance continuous spectrum of the present invention.
[0033] Figure 2 It is the rail continuous impedance spectrum-rail inductance continuous spectrum of the present invention. DETAILED DESCRIPTION
[0034] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0035] Example
[0036] In this embodiment, the rail impedance spectrum construction method based on ensemble learning considers that the rail impedance consists of two parts, resistance and inductance, and does not consider other electrical characteristics, and uses the correlation between the values of resistance and inductance and the frequency.
[0037] The specific steps of the construction method of this embodiment include:
[0038] S1. Construct and preprocess the dataset:
[0039] S1-1. Get the dataset:
[0040] A rail resistance and inductance dataset is obtained by field measurement, and the characteristics of the dataset include signal frequency, rail resistance and rail inductance;
[0041] The double short-circuit method can be used to measure rail parameters at railway application sites;
[0042] The measured rail section does not contain compensation capacitors, rail lead wires and other equipment.
[0043] S1-2. Preprocess the data set:
[0044] That is, outlier processing is performed on the data in the data set, which can be based on the 3σ criterion or the 6σ criterion. In this embodiment, based on the 3σ criterion, data points that exceed the judgment range of the 3σ criterion are judged as outliers;
[0045] The preprocessed dataset is divided into training set and test set.
[0046] S2. Construct resistance calculation model and inductance calculation model:
[0047] The rail resistance and inductance data sets were iteratively trained with the help of random forest algorithm and gradient boosting regression tree algorithm to obtain the resistance calculation model related to resistance and frequency and the inductance calculation model related to inductance and frequency.
[0048] The specific steps include:
[0049] S2-1. Constructing the computational model framework:
[0050] The rail equivalent impedance calculation model framework is constructed by an integrated learning algorithm. The constructed model framework is composed of a parallel and independent resistance calculation model and an inductance calculation model. The resistance calculation model and the inductance calculation model execute this step independently.
[0051] Before building the rail equivalent impedance calculation model framework, the hyperparameters are set. The hyperparameters in this implementation include the hyperparameters of the random forest algorithm and the hyperparameters of the gradient boosting regression tree algorithm.
[0052] Among them, the hyperparameters of the random forest algorithm include the number of weak learners, the maximum depth, and the maximum number of features; the hyperparameters of the gradient boosting regression tree algorithm include the learning rate, the number of weak learners, and the maximum depth.
[0053] S2-2. Construct rail impedance calculation model:
[0054] The training set is used to select and optimize the hyperparameter space in the ensemble learning model. The hyperparameter space selection is to pre-set the value range of the hyperparameter, and the hyperparameter space optimization is achieved through grid search and cross-validation.
[0055] A rail impedance calculation model is constructed, and the ensemble learning algorithms used include the random forest algorithm and the gradient boosting regression tree algorithm.
[0056] S2-3. Performance evaluation of rail impedance calculation model:
[0057] The test set is input into the constructed rail impedance calculation model, and the performance of the rail impedance calculation model is evaluated according to the calculation output results. The evaluation indicators are mean square error and determination coefficient:
[0058] The calculation formulas for mean square error and determination coefficient are as follows:
[0059]
[0060]
[0061] In formula 1 and formula 2, MSE represents mean square error, R 2 represents the coefficient of determination, n represents the number of samples in the test set, i represents the sequence number of the test set data, and y i Indicates the measured value, represents the model output value, Represents the average of the measured values of the test set.
[0062] When the value of the root mean square error is lower than 0.02 and the value of the determination coefficient is higher than 0.98, it is determined whether the evaluation result meets the requirements. If it meets the requirements, step S3 is executed; if not, step S2 is executed again;
[0063] The root mean square error and determination coefficient results in this embodiment are shown in Table 1.
[0064] Table 1: Root mean square error and determination coefficient results between measured and calculated values
[0065] MSE <![CDATA[R 2 ]]> resistance 0.014 0.988 inductance 0.012 0.991
[0066] It can be seen from the results in Table 1 that the value of the root mean square error is lower than the standard value, and the value of the determination coefficient is higher than the standard value, so the evaluation result meets the requirements, and step S3 can be executed.
[0067] S3. The required continuous frequency range is used as the input of the resistance calculation model and the inductance calculation model. The continuous rail impedance spectrum is obtained through calculation. The results are shown in Figure 1 and Figure 2 .
[0068] Application Examples
[0069] The rail impedance spectrum constructed by the present invention is used to calculate the rail resistance and rail inductance, and the measured values are compared with the calculated values at some typical frequencies. The results are shown in Table 2.
[0070] Table 2: Comparison of measured and calculated values for some typical frequencies
[0071]
[0072]
[0073] It can be seen from the results in Table 2 that, after comparison at some typical frequencies, it is found that the calculated value and the measured value of the rail impedance spectrum constructed by the present invention are very small in value, the average relative error of the rail resistance is less than 2.1%, and the average relative error of the rail inductance is less than 0.5%. It can be seen that the rail impedance spectrum constructed by the present invention is more scientific and accurate.
Claims
1. A rail impedance spectrum construction method based on ensemble learning, characterized in that: The rail impedance is composed of two parts: resistance and inductance. The construction method specifically includes the following steps: S1. Construct and preprocess the dataset: Obtain rail resistance and inductance data sets, preprocess the data sets, and divide the preprocessed data sets into a training set and a test set; S2. Construct resistance calculation model and inductance calculation model: The rail resistance and inductance data sets are iteratively trained using the random forest algorithm and the gradient boosting regression tree algorithm to obtain a resistance calculation model that associates resistance with frequency and an inductance calculation model that associates inductance with frequency. S3. The required continuous frequency range is used as the input of the resistance calculation model and the inductance calculation model, and a continuous rail impedance spectrum is obtained through calculation.
2. A rail impedance spectrum construction method based on ensemble learning according to claim 1, characterized in that: In the step S2, hyperparameters are set before building the resistance calculation model and the inductance calculation model. The hyperparameters include hyperparameters of the random forest algorithm and hyperparameters of the gradient boosting regression tree algorithm: The hyperparameters of the random forest algorithm include the number of weak learners, the maximum depth, and the maximum number of features; The hyperparameters of the gradient boosting regression tree algorithm include learning rate, number of weak learners and maximum depth.
3. The method for constructing rail impedance spectrum based on ensemble learning according to claim 2, characterized in that: The S2 step also includes: The hyperparameter space composed of the hyperparameters is selected and optimized by using the training set to obtain a rail impedance calculation model; the selection is to pre-set the value range of the hyperparameters, and the optimization is achieved by grid search and cross-validation; The test set is input into the resistance calculation model and the inductance calculation model, and the performance of the rail impedance calculation model is evaluated according to the calculation output results to determine whether the evaluation results meet the requirements. If not, S2 is executed again; if it meets the requirements, S3 is executed.
4. The method for constructing rail impedance spectrum based on ensemble learning according to claim 3, characterized in that: The performance evaluation indicators include root mean square error and determination coefficient, and the compliance requirement is that the root mean square error value is lower than 0.02 and the determination coefficient value is higher than 0.
98.
5. The method for constructing rail impedance spectrum based on ensemble learning according to claim 4, characterized in that: The mean square error and determination coefficient calculation formulas are as follows: In formula 1 and formula 2, MSE represents mean square error, R 2 represents the coefficient of determination, n represents the number of samples in the test set, i represents the sequence number of the test set data, and y i Indicates the measured value, represents the model output value, Represents the average of the measured values of the test set.
6. The method for constructing rail impedance spectrum based on ensemble learning according to claim 1, characterized in that: In the step S1, the characteristics of the data set include signal frequency, rail resistance and rail inductance; and the data set is obtained by field measurement.
7. A method for constructing rail impedance spectrum based on ensemble learning according to claim 6, characterized in that: The field measurement described above adopts the double short-circuit method to measure the rail parameters at the railway application site, and the measured rail section does not contain equipment such as compensation capacitors and rail lead wires.
8. The method for constructing rail impedance spectrum based on ensemble learning according to claim 1, characterized in that: In the step S1, the preprocessing is to perform outlier processing on the data in the data set.
9. The method for constructing rail impedance spectrum based on ensemble learning according to claim 1, characterized in that: The method for handling outliers is based on the 3σ criterion or the 6σ criterion, and data points that exceed the criterion judgment range are judged as outliers.