Track circuit fault prediction method and system based on kernel fuzzy clustering and CNN-NRBO-LSSVM

Abstract

The invention discloses a track circuit fault prediction method and system based on kernel fuzzy clustering and CNN-NRBO-LSSVM. The track circuit fault prediction method comprises the steps of S1, selecting monitoring parameters capable of obviously reflecting the state change trend of a track circuit, and analyzing the change trend of the monitoring parameters corresponding to different fault types; s2, acquiring track circuit monitoring data of the microcomputer monitoring system based on the monitoring parameters, preprocessing the monitoring data and dividing the monitoring data into a training sample and a test sample; s3, performing feature dimension reduction on the monitoring data through a principal component analysis method based on a kernel function, and dividing a track circuit performance state through a fuzzy clustering algorithm based on the kernel function; s4, inputting the degradation state data of the poor level and the fault state data of the fault level into a convolutional neural network, and extracting high-dimensional features of different fault types; and S5, constructing a least square support vector machine prediction model based on a Newton-Raphson intelligent optimization algorithm, realizing prediction of different fault types of the track circuit, and performing experimental verification.

Description

Technical Field

[0001] The present invention belongs to the technical field of track circuit fault prediction, and particularly relates to a track circuit fault prediction method and system based on kernel fuzzy clustering and CNN-NRBO-LSSVM. Background Art

[0002] As one of the important basic devices of the train operation control system, the ZPW-2000A insulated track circuit plays an important role in ensuring the safety and efficiency of railway transportation. The insulated track circuit is an important part of the ground equipment of the train control system for high-speed railways in China. Once a fault occurs, it will directly affect the railway transportation efficiency and even endanger the train operation safety. At present, the maintenance decision-making for track circuits still mainly relies on manual analysis, and it is necessary to manually screen and analyze the fault warning information obtained from the microcomputer monitoring system, which has problems such as high difficulty, low efficiency and easy error. Therefore, based on the track circuit monitoring data and combined with intelligent algorithms for fault prediction of the track circuit, it can provide scientific support for the accurate "condition-based maintenance" of the ZPW-2000A insulated track circuit. Summary of the Invention

[0003] In order to solve the technical problems existing in the background art, the purpose of the present invention is to provide a track circuit fault prediction method and system based on kernel fuzzy clustering and CNN-NRBO-LSSVM, so as to provide fault prediction for the ZPW-2000A insulated track circuit. The results show that this method can perform unsupervised clustering based on monitoring data, provide a more objective state division method from a data-driven perspective, and achieve accurate prediction of track circuit faults. Through comparative experiments, it is confirmed that this method can effectively avoid falling into local extrema, greatly improve the performance of the model, obtain a better solution at a lower computational cost, and provide a more scientific and intelligent method for track circuit fault prediction.

[0004] In order to solve the technical problems, the technical solution of the present invention is as follows:

[0005] A track circuit fault prediction method based on kernel fuzzy clustering and CNN-NRBO-LSSVM, the method comprising:

[0006] S1: Perform feature dimensionality reduction on the monitoring data by Kernel Principal Components Analysis (KPCA) based on a kernel function, and divide the performance degradation states of the track circuit by Kernel Fuzzy C-Means (KFCM) based on a kernel function, and divide the track circuit into: normal grade state, good grade state, poor grade state and fault grade state;

[0007] S2: Input the degraded state data of the poorer level and the fault state data of the fault level into a convolutional neural network to extract high-dimensional features of different fault types;

[0008] S3: Based on the extracted high-dimensional features, construct a least squares support vector machine (Newton-Raphson-based optimizer-Least Squares Support Vector Machines, abbreviated as NRBO-LSSVM) prediction model based on the Newton-Raphson intelligent optimization algorithm, that is, realize the prediction of different fault types of the track circuit.

[0009] Further, before the step S1, the method includes:

[0010] Determine the monitoring parameters that can significantly reflect the changing trend of the track circuit state, and analyze the changing trends of the monitoring parameters corresponding to different fault types;

[0011] Based on the determined monitoring parameters, obtain the track circuit monitoring data of the microcomputer monitoring system, preprocess the monitoring data, and divide the training samples and test samples.

[0012] Further, determining the monitoring parameters that can significantly reflect the changing trend of the track circuit state and analyzing the changing trends of the monitoring parameters corresponding to different fault types includes:

[0013] Analyze the operating principle of the ZPW-2000A track circuit, sort out the main fault types of the track circuit, select 8 voltage and current monitoring quantities that can reflect the changing state of the track circuit as the fault data feature set, process the collected track circuit fault data, analyze the relationship between the fault type and the fault symptom, and establish a fault type set.

[0014] Further, obtain the track circuit monitoring data of the microcomputer monitoring system for preprocessing, and use 70% of the processed samples as training samples and 30% of the samples as test samples.

[0015] Further, in the step S1, after the feature dimension reduction of the monitoring data is combined with KPCA, KFCM is used for clustering, the clustering effects when different values are taken for the clustering number c are evaluated through the fuzzy partition coefficient and the Xie-Beni index, and the performance degradation state of the track circuit is divided according to the clustering results combined with the on-site expert experience.

[0016] Further, in step S2, in order to extract the high-dimensional features of the poor-level status data and the fault status level data, a convolutional neural network is used, and its specific settings are as follows: 1 input layer (input data scale: 8*1*1), 1 convolutional layer (filters: 32, kernel_size: 3*1, activation: ReLU, padding:'same'), max pooling layer (pooling size: 2*1, Stride: 2*1), 1 convolutional layer (filters: 64, kernel_size: 3*1, activation: ReLU, padding:'same'), 1 fully connected layer, 1 output layer.

[0017] Further, in step S3, the specific steps of using the NRBO intelligent optimization algorithm to optimize the LSSVM parameters are as follows:

[0018] S301: Reconstruct the feature and label data obtained by the CNN into a new dataset for training the NRBO-LSSVM model, and re-partition it into training samples and validation samples;

[0019] S302: Initialize the parameters of NRBO and LSSVM;

[0020] S303: Input the optimization problem information in LSSVM into NRBO, and initialize the population. Continuously update the optimization parameters to be optimized, namely the penalty parameter C and the kernel parameter, through the positions of the population;

[0021] S304: Substitute the finally obtained hyperparameter combination into LSSVM for fault prediction, and then use the test samples to verify the accuracy of the model.

[0022] An orbit circuit fault prediction system based on kernel fuzzy clustering and CNN-NRBO-LSSVM, the system is applied to any one of the above methods, and the system includes:

[0023] A state monitoring parameter identification module, which is used to determine the monitoring parameters that can significantly reflect the change trend of the orbit circuit state, and analyze the change trends of the monitoring parameters corresponding to different fault types;

[0024] A data preprocessing and sample partitioning module, which is used to obtain the orbit circuit monitoring data of the microcomputer monitoring system based on the determined monitoring parameters, preprocess the monitoring data, and partition the training samples and test samples;

[0025] A feature dimension reduction and clustering module, which is used to perform feature dimension reduction on the monitoring data based on the principal component analysis method of kernel function, and divide the performance degradation state of the track circuit through the fuzzy clustering algorithm based on the kernel function. The track circuit is divided into: normal level state, good level state, poor level state and fault level state;

[0026] A feature extraction module, which is used to input the degradation state data of the poor level and the fault state data of the fault level into the convolutional neural network to extract the high-dimensional features of different fault types;

[0027] A fault prediction model construction module, which is used to construct a least squares support vector machine prediction model based on the Newton-Raphson intelligent optimization algorithm based on the extracted high-dimensional features, that is, to realize the prediction of different fault types of the track circuit.

[0028] A computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a track circuit fault prediction method based on kernel fuzzy clustering and CNN-NRBO-LSSVM as described in any one of the above.

[0029] A computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements a track circuit fault prediction method based on kernel fuzzy clustering and CNN-NRBO-LSSVM as described in any one of the above.

[0030] Compared with the prior art, the advantages of the present invention are as follows:

[0031] (1) There is a transient to steady-state process in the state change of the track circuit. KFCM fuzzy clustering can perform unsupervised clustering based on the monitoring data, providing a more objective state division method from the perspective of data-driven.

[0032] (2) CNN extracts target features from the original degradation data, obtains the feature changes of the track circuit, and combines with the NRBO-LSSVM prediction model to accurately realize the track circuit fault prediction.

[0033] (3) The prediction accuracy of CNN-NRBO-LSSVM reaches 98.67%. Compared with the traditional support vector machine, the prediction accuracy has been significantly improved. In addition, through comparative experiments, it is confirmed that the NRBO optimization algorithm can effectively avoid falling into local extrema, greatly improve the performance of the model, obtain a better solution with lower computational cost, and provide a more scientific and intelligent method for track circuit fault prediction. Description of the Drawings

[0034] Figure 1 It is the overall process framework diagram for fault prediction;

[0035] Figure 2 is the KFCM contour clustering diagram ( Figure 2a for c = 3, Figure 2b for c = 4, Figure 2c for c = 5);

[0036] Figure 3 is the block diagram of the NRBO-LSSVM algorithm;

[0037] Figure 4 is the model training process diagram ( Figure 4a is the change in accuracy, Figure 4b is the change in loss iteration);

[0038] Figure 5 is the experimental prediction result diagram ( Figure 5a is the prediction result of the test set, Figure 5b is the confusion matrix of the test set, Figure 5c is the visualization of output features);

[0039] Figure 6 is the process diagram of iterative optimization of different optimization algorithms. Specific implementation mode

[0040] The following describes the specific implementation mode of the present invention in combination with embodiments:

[0041] It should be noted that the structures, ratios, sizes, etc. shown in this specification are only used to cooperate with the content disclosed in the specification for those familiar with this technology to understand and read, and are not used to limit the implementation conditions of the present invention. Any modification of the structure, change in the ratio relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope covered by the technical content disclosed in the present invention.

[0042] At the same time, the terms such as "upper", "lower", "left", "right", "middle", and "one" cited in this specification are only for the convenience of clear description and are not used to limit the scope of implementation of the present invention. The change or adjustment of their relative relationship, without substantial change in the technical content, should also be regarded as the scope within which the present invention can be implemented.

[0043] Embodiment 1:

[0044] As Figure 1 shown, a track circuit fault prediction method based on kernel fuzzy clustering and CNN-NRBO-LSSVM includes the following steps:

[0045] S1. Select monitoring parameters that can significantly reflect the changing trend of the track circuit state and analyze the changing trends of the monitoring parameters corresponding to different fault types.

[0046] At present, the railway department mainly uses the ZPW-2000A of the microcomputer monitoring system to realize the monitoring of switch quantities and analog quantities. Among them, the equipment voltage and current values included in the analog quantities will show different fluctuations with the change of the operating state. Referring to documents such as TG / XH101-2015 "Maintenance Rules for Signals of Regular Speed Railways" and "Technical Standards for ZPW-2000 Track Circuits", combined with the acquisition content of the microcomputer monitoring system and the parameters in the actual monitoring data of the electric signal section that can reflect the operation trend of the track circuit, 8 monitoring parameters are selected as the input of the data set, as shown in Table 1.

[0047] Table 1 - Monitoring Parameters

[0048]

[0049] The essence of the ZPW-2000A track circuit failure lies in the sudden change of the track circuit parameters, resulting in the dynamic change process of the track circuit signal amplitude from the transient state to the steady state. When a specific device of the track circuit is about to fail, there are often multiple fault symptoms. For example, a certain voltage value will continue to rise until it exceeds the limit, and one fault symptom may correspond to multiple fault types. Based on different parameter change trends, the fault type can be accurately determined. Based on this, according to the actual on-site alarm and maintenance records and combined with the track circuit fault diagnosis materials, the fault types and fault symptoms are given, as shown in Table 2.

[0050] Table 2 - Table of Track Circuit Fault Types and Fault Symptoms

[0051]

[0052] Note: "On the low side" means that the monitoring data is significantly lower than the normal state data or close to the alarm lower limit; "On the high side" means that the monitoring data is significantly higher than the normal state data or close to the alarm upper limit; "Decrease" means that the monitoring data continues to decrease and exceeds the alarm lower limit; "Increase" means that the monitoring data continues to rise and exceeds the alarm upper limit.

[0053] S2. Obtain the track circuit monitoring data of the microcomputer monitoring system, preprocess the monitoring data of the selected monitoring parameters, and divide the training set and the test set.

[0054] The original data of this embodiment comes from the ZPW-2000A track circuit monitoring data collected by a certain electric signal section of the Chengdu Railway Bureau, and the fault types are marked according to the corresponding maintenance alarm records. The period for the microcomputer monitoring system to collect the track circuit analog quantity is 250 ms, and 240 are collected per minute.

[0055] First, by analyzing the maintenance alarm records, determine the specific occurrence time of the fault. Based on this time node, extract various operation parameter data within 72 hours before the fault occurs. The data collection interval is set to 1 hour, and within each hour, calculate the average value of the 1-minute data in this time period to reduce the impact of short-term fluctuations on data analysis and ensure the smoothness and representativeness of the obtained data. Second, for the data processing after the fault occurs, the collection time period lasts from the moment the fault occurs until the fault is restored. In this time interval, extract and calculate the average value of the data every 1 minute at a fixed time interval of 1 minute. In this way, the dynamic changes of various parameters during the fault occurrence process can be reflected. Finally, through the above data processing method, the fault monitoring samples of the ZPW-2000A track circuit at a specific moment are obtained, as shown in Table 3.

[0056] Table 3 - Track Circuit Monitoring Data Table

[0057]

[0058] Considering that the dimensions of the monitoring parameters are not completely consistent, further standardization processing is required, and its formula is Formula (1).

[0059] S3. Use KPCA for feature dimensionality reduction of the monitoring data, and divide the performance state of the track circuit through the KFCM algorithm. Its basic steps are as follows:

[0060] S301. Assume that the dataset in the input space is X = {x1, x2, …, xn}, where n is the number of samples; xn = {xn1, xn2, …, xnk}, where k is the characteristic attribute of each sample; assume that the dataset is divided into c categories, and the clustering center of each category is represented as V = {v1, v2, …, vc}; the membership matrix of the data samples is μab (a = 1, 2, …, c; b = 1, 2, …, n), which represents the membership value of the b-th sample belonging to the a-th cluster.

[0061] S302. The essence of KFCM is to find the maximum μab for each sample point. The closer the sample is to the clustering center, the larger μab is. Therefore, model the minimum target distance, and the formula is Equation (2). Based on the data characteristics, this paper uses the Gaussian kernel function, and its formula is Equation (3).

[0062] S303. Use the necessary condition for the extreme value of Lagrange to derive the iterative formulas (4) of U and V according to Formulas (2) and (3).

[0063] S304. Set the number of clusters, hyperparameter m, and convergence accuracy δ, let the number of iterations k = 0, and initialize the clustering center V(0).

[0064] S305. Let \(k = k + 1\), and update the cluster center \(V(k + 1)\) and the membership matrix \(U(k + 1)\) according to Equation (4).

[0065] S306. Repeat S305 until \(\|U\) (k+1) -U (k) \(\|\leq\delta\).

[0066] S307. Evaluate the clustering effect for different values of \(c\) through the Fuzzy Partition Coefficient (FPC) and the Xie - Beni Index (XB), and divide the performance degradation state of the track circuit according to the clustering results combined with on - site expert experience.

[0067] In the present invention, the eight monitoring parameters of the track circuit are dimension - reduced by KPCA. According to expert experience, the track circuit can be roughly divided into 3 - 5 states from normal to faulty for unsupervised clustering. The KFCM parameters are set as: \(m = 2,\ \epsilon=1e - 06,\ \beta = 0.5\). The clustering diagram is shown in Figure 2. It can be seen from Figure 2 that some discrete points in the figure are mutation data caused by faulty states. When \(c = 4\), the greater the membership degree of the sample to the cluster, the clearer and better the clustering effect. The results of its evaluation indexes are shown in Table 4.

[0068] Table 4 - Quantitative evaluation results of different clusterings

[0069]

[0070] It can be obtained from Table 5 that when the number of clusters \(c = 4\), the closer the FPC is to 1 and the closer the XB is to 0, the better the clustering effect. Combining the actual on - site situation, the track circuit is divided into 4 grades from normal to faulty, as shown in Table 5.

[0071] Table 5 - State division of the track circuit

[0072]

[0073] S4. Input the data of the degradation states of the poorer grades and the faulty state data of the faulty grades into the Convolutional Neural Network (CNN) to extract high - dimensional features of different fault types.

[0074] The hyperparameters of the CNN model used in this article are obtained through previous experience and comparative experiments. The CNN model consists of an input layer, a feature layer (two convolutional layers and one pooling layer), an FC layer, and an output layer. The SGDM algorithm is used as the optimization algorithm, and its specific parameters are shown in Table 6.

[0075] Table 6 - Model hyperparameter settings of the CNN model

[0076]

[0077] S5. Build an NRBO-LSSVM prediction model to achieve the prediction of different fault types of track circuits and conduct experimental verification.

[0078] Figure 3 Figure 4 is the block diagram of the NRBO-LSSVM algorithm. The specific steps of using the NRBO intelligent optimization algorithm to optimize the LSSVM parameters are as follows:

[0079] S501. Reconstruct the feature and label data obtained by CNN into a new dataset for NRBO-LSSVM model training, and re-divide it into training samples and validation samples.

[0080] In the present invention, the classification accuracy of training samples and validation samples is used as the evaluation index of the model performance. The maximum number of iterations is set to 100, and the CNN-NRBO-LSSVM model is trained. Its training process is shown in Figure 4. During the model training process, as the number of iterations gradually increases, when the training reaches about 50 times, the performance of the model begins to tend to be stable. At this time, the accuracy rate steadily increases and finally reaches about 98%, indicating that the prediction ability of the model on the training samples has been significantly improved. At the same time, the loss value is close to zero, which means that the error of the model has been reduced to the minimum level, and the fitting effect of the model tends to be optimal. This result shows that the training process of the model on this task has tended to converge, can effectively capture the potential patterns in the data, and has achieved good prediction results.

[0081] S502. Initialize the parameters of NRBO and LSSVM;

[0082] S503. Input the optimization problem information in LSSVM into NRBO, and initialize the population. Continuously update the optimization parameters required, that is, the penalty parameter C and the kernel parameter, through the positions of the population;

[0083] S504. Substitute the finally obtained hyperparameter combination into LSSVM for fault prediction, and then use the test samples to verify the accuracy of the model.

[0084] In order to further verify the accuracy and actual application effect of the prediction model, the degraded data of the lower grade in the test samples is selected and input into the trained prediction model for evaluation. The obtained prediction results are shown in Figure 5.

[0085] As can be seen from Figure 5, among the 448 test samples, six samples were misclassified into other categories. The confusion matrix shown in Figure 5(b) is used to evaluate the performance of the prediction model in the task of track circuit fault prediction. The horizontal and vertical coordinates of the confusion matrix correspond to the 8 fault types (F1 - F8) of the track circuit from 1 to 8 respectively. The values on the diagonal of the matrix represent the correct prediction quantity of the model for the degraded samples of each category. By analyzing the confusion matrix, the classification effect of the model on different fault categories can be intuitively understood.

[0086] As can be seen from Figure 5, for the 4 fault types of F1, F2, F4, and F6, the model achieved 100% accurate prediction, that is, all test samples were correctly classified into the corresponding categories through the prediction model, indicating that the model has extremely high robustness for these fault types. For the remaining 4 fault types (F3, F5, F7, F8), the classification accuracy rates of the model reached 96.6%, 98.3%, 96.9%, and 98.5% respectively. Although there are a small number of errors, these results still show that the model has good recognition ability for relatively complex fault types, and the prediction accuracy remains at a high level.

[0087] In this embodiment, in order to further verify the superiority of the model, multiple different prediction models were evaluated and compared using the same test samples. The specific comparison methods selected were: SVM, LSSVM, NRBO - LSSVM, CNN - NRBO - LSSVM. The comparison results are shown in Table 7:

[0088] Table 7 - Comparison Results of Different Prediction Models

[0089]

[0090] As can be seen from Table 7, LSSVM has greatly improved in prediction time compared with SVM, verifying the significant improvement in computational cost of LSSVM. CNN - NRBO - LSSVM has a relatively large improvement in accuracy compared with NRBO - LSSVM. The data of the degraded state of the lower - grade test samples and the fault - state data of the fault grade in the training samples were input into the convolutional neural network (CNN), and feature dimensionality reduction and visualization were performed through t - distributed stochastic neighbor embedding (t - SNE), as shown in Figure 5(c), which confirmed the powerful feature extraction ability of CNN. NRBO - LSSVM also has a relatively large improvement in accuracy compared with LSSVM, confirming that the intelligent algorithm for determining the values of the LSSVM hyperparameters is beneficial to improving the accuracy of the model.

[0091] In this embodiment, to further illustrate the improvement of the NRBO intelligent algorithm on the model performance, the Sparrow Search Algorithm (SSA), Natural Gradient Optimization Algorithm (NGO), Particle Swarm Optimization (PSO), and Whale Optimization Algorithm (WOA) are applied to the determination of the hyperparameters of LSSVM and compared with NRBO-LSSVM. In the above experiments, the population size is set to 20, the maximum number of iterations is 10, and the value ranges of C and σ 2 are (0.001, 100), where the inertia factor w of PSO is 0.9, and the acceleration constants c1 and c2 are equal to 2. The fitness curves are as Figure 6 shown, and the model prediction results are shown in Table 8.

[0092] Table 8 - Prediction Results of LSSVM Optimized by Different Optimization Algorithms

[0093]

[0094] As can be seen from Table 8, the accuracy of the prediction model optimized by NRBO is 2.42%, 3.43%, 1.52%, and 5.22% higher than the models optimized by SSA, NGO, WOA, and PSO, respectively. Analysis Figure 6 , it can be found that as the number of iterations increases, the fitness of NRBO drops rapidly, and it has a more obvious population evolution advantage compared with other optimization algorithms. Analyzing the reasons, it can be seen that the unique trap avoidance operation of NRBO can make the population more diverse and escape from the local optimal solution, and its global search ability is significantly stronger than other optimization algorithms.

[0095] Table 9 - Formulas and Explanations Used in the Invention

[0096]

[0097] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0098] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks

[0099] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks

[0101] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those of ordinary skill in the art

[0102] Many other changes and modifications can be made without departing from the concept and scope of the present invention. It should be understood that the present invention is not limited to specific embodiments, and the scope of the present invention is defined by the appended claims

Claims

1. A method for predicting track circuit faults based on kernel fuzzy clustering and CNN-NRBO-LSSVM, characterized in that, The method includes: S1: Perform feature dimensionality reduction on the monitoring data based on the kernel principal component analysis method, and divide the performance degradation state of the track circuit through the fuzzy clustering algorithm based on the kernel function. The track circuit is divided into: normal level state, good level state, poor level state, and fault level state; S2: Input the degradation state data of the poor level and the fault state data of the fault level into the convolutional neural network to extract high-dimensional features of different fault types; S3: Based on the extracted high-dimensional features, construct a least squares support vector machine prediction model based on the Newton-Raphson intelligent optimization algorithm, that is, realize the prediction of different fault types of the track circuit.

2. The orbit circuit fault prediction method based on kernel fuzzy clustering and CNN-NRBO-LSSVM according to claim 1, wherein, Before the step S1, the method includes: Determine the monitoring parameters that can significantly reflect the change trend of the track circuit state, and analyze the change trends of the monitoring parameters corresponding to different fault types; Based on the determined monitoring parameters, obtain the track circuit monitoring data of the microcomputer monitoring system, preprocess the monitoring data, and divide the training samples and test samples.

3. A method for predicting track circuit faults based on kernel fuzzy clustering and CNN-NRBO-LSSVM according to claim 2, characterized in that, Determine the monitoring parameters that can significantly reflect the change trend of the track circuit state, and analyze the change trends of the monitoring parameters corresponding to different fault types, including: Analyze the operating principle of the ZPW-2000A track circuit, sort out the main fault types of the track circuit, select 8 voltage and current monitoring quantities that can reflect the change of the track circuit state as the fault data feature set, process the collected track circuit fault data, and analyze the relationship between the fault type and the fault symptom, and establish a fault type set.

4. A method for predicting rail circuit faults based on kernel fuzzy clustering and CNN-NRBO-LSSVM according to claim 1, characterized in that, Obtain the track circuit monitoring data of the microcomputer monitoring system for preprocessing, and use 70% of the processed samples as training samples and 30% of the samples as test samples.

5. A method for predicting track circuit faults based on kernel fuzzy clustering and CNN-NRBO-LSSVM according to claim 1, characterized in that, In the step S1, after performing feature dimensionality reduction on the monitoring data combined with KPCA, use KFCM for clustering, evaluate the clustering effect when different values are taken for the clustering number c through the fuzzy partition coefficient and the Xie-Beni index, and divide the performance degradation state of the track circuit according to the clustering results combined with the on-site expert experience.

6. A method for predicting track circuit faults based on kernel fuzzy clustering and CNN-NRBO-LSSVM according to claim 1, characterized in that, In the step S2, in order to extract the high-dimensional features of the poor level state data and the fault state level data, use a convolutional neural network, and its specific settings are: 1 input layer (input data scale: 8*1*1), 1 convolutional layer (filters: 32, kernel_size: 3*1, activation: ReLU, padding:'same'), max pooling layer (pooling size: 2*1, Stride: 2*1), 1 convolutional layer (filters: 64, kernel_size: 3*1, activation: ReLU, padding:'same'), 1 fully connected layer, 1 output layer.

7. A method for predicting track circuit faults based on kernel fuzzy clustering and CNN-NRBO-LSSVM according to claim 1, characterized in that, In the step S3, the specific steps of using the NRBO intelligent optimization algorithm to optimize the LSSVM parameters are as follows: S301: Reconstruct the features and label data obtained by the CNN into a new data set for training the NRBO-LSSVM model, and re-divide it into training samples and validation samples; S302: Initialize the parameters of NRBO and LSSVM; S303: Input the optimization problem information in LSSVM into NRBO, initialize the population, and continuously update the optimization parameters to be determined, namely the penalty parameter C and the kernel parameter, through the positions of the population; S304: Substitute the finally obtained hyperparameter combination into LSSVM for fault prediction, and then verify the accuracy of the model with test samples.

8. An orbit circuit fault prediction system based on kernel fuzzy clustering and CNN-NRBO-LSSVM, characterized in that, The system is applied to the method described in any one of claims 1-7, and the system includes: A state monitoring parameter identification module, configured to determine monitoring parameters that can significantly reflect the changing trend of the track circuit state, and analyze the changing trends of the monitoring parameters corresponding to different fault types; A data preprocessing and sample division module, configured to obtain the track circuit monitoring data of the microcomputer monitoring system based on the determined monitoring parameters, preprocess the monitoring data, and divide the training samples and test samples; A feature dimensionality reduction and clustering module, configured to perform feature dimensionality reduction on the monitoring data based on the principal component analysis method of the kernel function, divide the performance degradation states of the track circuit through the fuzzy clustering algorithm based on the kernel function, and divide the track circuit into: normal grade state, good grade state, poor grade state, and fault grade state; A feature extraction module, configured to input the degradation state data of the poor grade and the fault state data of the fault grade into a convolutional neural network to extract high-dimensional features of different fault types; A fault prediction model construction module, configured to construct a least squares support vector machine prediction model based on the Newton-Raphson intelligent optimization algorithm based on the extracted high-dimensional features, that is, to realize the prediction of different fault types of the track circuit.

9. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a track circuit fault prediction method based on kernel fuzzy clustering and CNN-NRBO-LSSVM described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the program is executed by the processor, it implements a track circuit fault prediction method based on kernel fuzzy clustering and CNN-NRBO-LSSVM described in any one of claims 1 to 7.

Images