ZPW-2000 type track circuit shunt bad fault trend early warning algorithm
By combining onboard and ground signals and using a bidirectional LSTM neural network model, the problem of early warning for faulty shunt circuits in the ZPW-2000 track circuit was solved, enabling accurate identification and location of faulty shunt trends and reducing safety hazards.
Patent Information
- Application Number
- CN202512031545.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-30
- Publication Date
- 2026-02-10
AI Technical Summary
Existing technologies cannot effectively prevent shunt failures in the ZPW-2000 track circuit, which can lead to safety hazards. Existing early warning methods can only provide information after a failure occurs and cannot provide early warning.
By acquiring onboard and ground monitoring signals, segmentation processing is performed using the carrier frequency cross-configuration characteristics of adjacent track circuits in the section. Combined with sliding window movement and cubic spline interpolation, time-aligned signals are generated and input into a bidirectional LSTM neural network to train a shunt failure trend early warning model. By combining local location shunt status prediction and compensation capacitor information, the shunt failure trend is located.
It enables trend-based early warning of shunt malfunctions, improving the accuracy and practicality of the warnings. It can identify the shunt status at local locations before a fault occurs, thus reducing safety risks.
Smart Images
Figure CN121493040A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of rail transit technology, specifically to a ZPW-2000 type track circuit shunt failure trend early warning algorithm. Background Technology
[0002] The ZPW-2000 track circuit is a crucial piece of railway signaling infrastructure, responsible for checking track occupancy and transmitting train operation information. When a train enters a section, the circuit formed by the wheels and rails short-circuits the transmitter signal, causing a drop in rail-out voltage and triggering the track relay, indicating that the section is occupied. However, if the wheels or rails are rusted, causing the wheel-rail shunt resistance to exceed the standard shunt sensitivity, the rail-out voltage upon train entry may still be higher than normal, preventing the track relay from triggering properly and resulting in a shunt malfunction. A shunt malfunction will cause the track circuit's train occupancy check to fail, leading to the loss of section status information and potentially causing a major safety accident.
[0003] Currently, railway sites are avoiding circuit malfunctions by upgrading existing track circuits to 3V, implementing high-voltage pulse asymmetry, and installing axle counting equipment. However, due to factors such as construction difficulty and project cost, these methods cannot completely solve the problem of circuit malfunctions, and there is an urgent need for low-cost, effective, and intelligent solutions.
[0004] Existing research primarily relies on estimating shunt resistance using locomotive signals or identifying shunt malfunctions based on rail output voltage. While locomotive signals can predict shunt resistance, they lack information on the operational status of ground-based systems, thus failing to effectively diagnose shunt malfunctions. Rail output voltage data collected by ground monitoring systems can identify shunt malfunctions, but these methods only provide information after a shunt malfunction has occurred and cannot prevent it from happening in the first place. Summary of the Invention
[0005] To address the aforementioned shortcomings in the existing technology, this invention provides a ZPW-2000 track circuit shunt failure trend early warning algorithm to solve the problem of inaccurate shunt failure judgment and early warning in existing shunt failure early warning methods.
[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A fault trend early warning algorithm for poor shunt circuits in ZPW-2000 track circuits includes the following steps: S1. Acquire vehicle-mounted monitoring signals and ground monitoring signals; S2. Based on the carrier frequency crossover configuration characteristics of adjacent track circuits in the interval, the on-board monitoring signal is segmented to generate segmented on-board monitoring signals. S3. Move the sliding window of the segmented onboard monitoring signals to generate the main track signal; S4. Acquire the ground monitoring signal from the main track signal from the start time to the set moving time threshold, move the sliding window, and generate the time-aligned vehicle monitoring signal and ground monitoring signal. S5. Perform cubic spline interpolation and resampling on the time-aligned vehicle-mounted monitoring signal and ground monitoring signal respectively to generate resampled vehicle-mounted monitoring signal and ground monitoring signal. S6. Obtain the onboard monitoring signal and ground monitoring signal of any track section. After executing steps S1-S5, input them into a bidirectional LSTM neural network for training to generate a fault trend early warning model for poor routing. S7. Obtain the onboard monitoring signal and ground monitoring signal of the track section to be tested. After executing steps S1-S5, input the shunt failure trend early warning model to predict the local shunt status. At the same time, introduce the peak method to obtain the compensation capacitor information of the track section to be tested. Combine the local shunt status prediction results to locate the shunt failure trend location.
[0007] The present invention has the following beneficial effects: 1. The ZPW-2000 type track circuit shunt failure trend early warning algorithm proposed in this invention combines on-board monitoring signals and ground monitoring signals, and establishes a shunt failure trend early warning model to identify the shunt status at local locations, thereby providing shunt failure trend early warning information. While realizing vehicle-ground data synchronization and alignment and local shunt failure identification and early warning, it has higher practicality. 2. This invention also considers onboard monitoring signals and ground monitoring signals, which are less affected by changes in line parameters and the state of compensation capacitors. This results in higher accuracy than simply using locomotive signals to estimate shunt resistance or relying on the magnitude of the rail output voltage from ground monitoring signals to identify shunt defects. Attached Figure Description
[0008] Figure 1 This is a flowchart illustrating a fault trend early warning algorithm for a ZPW-2000 type track circuit shunt failure proposed in this invention. Figure 2 This is a schematic diagram of the data before and after segmentation processing of the vehicle-mounted monitoring signal in section 1412BG in the embodiment; Figure 3 This is a data diagram illustrating the generation of the main track signal by moving the sliding window of the on-board monitoring signal in section 1412BG of the embodiment. Figure 4 This is a schematic diagram of the original data of the onboard and ground monitoring signals of the track circuit in section 1710BG of the embodiment; Figure 5 This is a schematic diagram of the data after time alignment of the on-board and ground monitoring signals of the track circuit in section 1710BG in the embodiment; Figure 6 This is a schematic diagram of the data after aligning the lengths of the onboard and ground monitoring signals for the track circuit in section 1710BG of the embodiment; Figure 7 This is a schematic diagram of the bidirectional LSTM neural network in the embodiment; Figure 8 This is a schematic diagram illustrating the structural principle of a single LSTM cell unit in the first or second bidirectional LSTM neural network layer in the embodiment. Figure 9 This is a schematic diagram of the identification results of the compensation capacitor position information of the ground monitoring signal of the track circuit in section 1710BG in the embodiment; Figure 10 This is a schematic diagram of the experimental results for early warning of poor track circuit shunt trend in section 1430AG of the embodiment; Figure 11 This is a schematic diagram of the experimental results for early warning of poor track circuit shunt trend in section 1568BG of the embodiment; Figure 12 This is a schematic diagram of the macro-F1 score experimental results when training the poor routing trend early warning model in the example; Figure 13 This is a schematic diagram of the confusion matrix experimental results when training the poor routing trend early warning model in the example. Detailed Implementation
[0009] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0010] like Figure 1 As shown, a fault trend early warning algorithm for poor shunt circuits in the ZPW-2000 type track circuit includes the following steps S1-S7: S1. Acquire vehicle-mounted monitoring signals and ground monitoring signals.
[0011] In this embodiment, the on-board monitoring signal is the monitoring information of the track inspection vehicle or the locomotive signal information, including information such as time, signal, carrier frequency, induced voltage and compensation capacitor induced pulse; the ground monitoring signal is the track circuit monitoring information collected by the microcomputer monitoring system or the centralized signal monitoring system, including information such as power output voltage, power output current, voltage on the sending end cable side, current on the sending end cable side, main rail voltage on the receiving end cable side and main rail voltage at the receiving entrance.
[0012] S2. Based on the carrier frequency crossover configuration characteristics of adjacent track circuits in the section, the on-board monitoring signal is segmented to generate segmented on-board monitoring signals.
[0013] In this embodiment, the carrier frequency cross-configuration characteristic of adjacent track circuits in the section refers to the fact that the carrier frequencies of adjacent track circuit sections are typically cross-configured at 1700Hz and 2300Hz or 2000Hz and 2600Hz. The process of segmenting the onboard monitoring signal is as follows: Starting from the starting point of the vehicle-mounted monitoring signal, the carrier frequency is recorded along the line direction and the points of change of the carrier frequency are found. Based on the points of change of the carrier frequency, the first point of change of the carrier frequency is taken as the starting point of the current track segment and the ending point of the previous track segment, and the second point of change of the carrier frequency is taken as the ending point of the current track segment and the starting point of the next track segment, and so on, so as to finally realize the segmentation of the vehicle-mounted monitoring signal.
[0014] In this embodiment, the starting point of the onboard monitoring signal is determined according to the principle of cross-configuration of track circuit carrier frequencies: the downline uses the 1700Hz and 2300Hz series; the upline uses the 2000Hz and 2600Hz series. The signal starting point is determined by the carrier frequency abrupt change point. For example, when the train travels from 1690AG (1700Hz) through 1692BG (2300Hz) to 1694CG (1700Hz), the carrier frequency data of the onboard monitoring signal shows abrupt changes from 1700Hz to 2300Hz and from 2300Hz to 1700Hz. These two points correspond to the start and end times of the 1692BG onboard monitoring signal, respectively. Similarly, the determination of the change point is the same as above. Because the locomotive monitoring signal analysis carrier frequency is sufficiently sensitive and accurate, it can accurately detect changes in carrier frequency such as from 1700Hz to 2300Hz and from 1700Hz to 2300Hz. In the experiment, the second point in a 300Hz frequency difference between two adjacent points is set as the change point. Figure 2 The data before and after segmented processing of the on-board monitoring signal in section 1412BG is displayed.
[0015] S3. Move the sliding window of the segmented onboard monitoring signals to generate the main track signal.
[0016] In this embodiment, the segmented on-board monitoring signal includes two parts: the main track signal and the secondary track signal. Furthermore, there is a sudden drop in induced voltage at the boundary between the main track and secondary track signals. Therefore, based on this characteristic, a sliding window method is used to extract the main track signal from the segmented on-board monitoring signal. The process is as follows: Set size to The sliding window, based on the induced voltage data in the segmented vehicle monitoring signal, calculates adjacent values sequentially from the receiver to the transmitter. The variance between each point (i.e., the induced voltage data point) is used to construct the first variance array. .
[0017] In this embodiment, under normal circumstances, the train runs in the forward direction from the receiving end to the sending end. Therefore, the starting data corresponds to the train position at the receiving end, and the ending data corresponds to the train position at the sending end. Furthermore, this invention only considers the scenario of the train running in the forward direction.
[0018] For the first variance array A traversal search is performed, and the first point where the variance changes drastically is taken as the starting point of the induced voltage data. At the same time, the second point where the variance changes drastically is taken as the ending point of the induced voltage data. Finally, the induced voltage data from the starting point to the ending point is taken as the main track signal.
[0019] In this embodiment, the induced voltage is almost zero in the small track section, such as Figure 2 The 1412BG segment exhibits a rapidly decreasing waveform before and after it, corresponding to points of drastic variance change. Therefore, in the experiment, the points with the largest variance in the first and second halves of the waveform were selected as the points of change, and these were considered as points of drastic variance change. Figure 3 The data results of generating the main track signal after the sliding window of the on-board monitoring signal in section 1412BG is shown.
[0020] S4. Acquire the ground monitoring signal from the main track signal from the start time to the set moving time threshold, move the sliding window, and generate time-aligned vehicle monitoring signal and ground monitoring signal.
[0021] In this embodiment, ground monitoring signals are acquired within the period from the start time of the main track signal to a set movement time threshold (the movement time threshold is 5 minutes). The starting and ending points of the main track pressure change at the receiving inlet of the ground monitoring signal are then searched based on variance characteristics to achieve time alignment between the monitoring signal and the ground monitoring signal. The process is as follows: Acquire ground monitoring signals from the main orbit signal from the start time to a set movement time threshold.
[0022] Set size to The sliding window, based on the main rail voltage data received from the ground monitoring signal reception entrance, calculates adjacent values sequentially from the starting point to the ending point. Construct a second variance array based on the variance between points. .
[0023] In this embodiment, the start time (start point) of the ground monitoring signal is 5 minutes before the start of the main track signal, and the end time (end point) is 5 minutes after the end of the main track signal. Due to the asynchrony between the onboard clock and the ground clock, there is an error in the time field recorded by the onboard monitoring signal and the ground monitoring signal when the train passes through a certain section (the error is within the set time threshold range). The ground monitoring signal data obtained by this method has a larger range, including valid data (train passed) and invalid data (train did not pass), and further processing is required.
[0024] For the second variance array A traversal search is performed, and the first point where the variance changes drastically is taken as the starting point of the train entering the track section, and the second point where the variance changes drastically is taken as the ending point of the train leaving the track section, thus obtaining the time of the starting point and the ending point.
[0025] In this embodiment, this step is to extract effective ground monitoring signals; after the train enters the track section, the rail outlet voltage drops rapidly to about 0V and remains stable until the train leaves the track, at which point the rail outlet voltage quickly recovers to the level before the train entered; therefore, the points of change when the train enters and leaves the track correspond to the points of drastic change in variance; therefore, in the experiment, the points with the largest variance in the first half and the second half of the waveform are selected as the points of change, and these are taken as the points of drastic change in variance.
[0026] Ground monitoring data is extracted from the time period from the start point to the end point to achieve time alignment between vehicle-mounted monitoring signals and ground monitoring signals.
[0027] In this embodiment, Figure 4 The raw data of onboard and ground monitoring signals for the track circuits in section 1710BG are displayed. Figure 5 The data shows the time-aligned data of the onboard and ground monitoring signals for the track circuit in section 1710BG.
[0028] S5. Perform cubic spline interpolation and resampling on the time-aligned vehicle-mounted monitoring signal and ground monitoring signal respectively to generate resampled vehicle-mounted monitoring signal and ground monitoring signal.
[0029] Specifically, step S5 includes: A set of discrete data points is used to represent time-aligned vehicle-mounted or ground-based monitoring signals, i.e.:
[0030] in, , , , , They represent the 1st, 2nd, and 3rd respectively. The, the The, the The time for each discrete data point , , , , They represent the 1st, 2nd, and 3rd respectively. The, the The, the The values of discrete data points correspond to the voltage, current, and other values of the corresponding signal.
[0031] For each pair of adjacent discrete data points and discrete data points Using cubic interpolation polynomials Describe the adjacent interval The cubic interpolation, i.e.:
[0032] in, Indicates adjacent intervals The cubic interpolation polynomial, , , , Each represents an unknown coefficient to be solved. Indicates the time of the original data.
[0033] Set a cubic interpolation polynomial The second derivative at the boundary is zero, and for every pair of adjacent discrete data points... and discrete data points cubic interpolation polynomial All satisfy the condition that the cubic interpolation polynomials of adjacent intervals are equal at the data points, the first derivative, and the second derivative, so as to generate a system of linear equations containing unknown coefficients.
[0034] The Gaussian elimination method is used to solve the system of linear equations containing unknown coefficients, ultimately generating the cubic interpolation polynomial for the solution. .
[0035] By splicing together the cubic interpolation polynomials of each adjacent interval, the fitting curve function of the discrete data points is obtained, which is the fitted vehicle monitoring signal or ground monitoring signal, thus realizing the cubic spline interpolation of the time-aligned vehicle monitoring signal or ground monitoring signal.
[0036] The fitted vehicle-mounted monitoring signal or ground monitoring signal is sampled at equal intervals, and the number of sampling points is twice the length of the ground monitoring signal before fitting, to generate a resampled vehicle-mounted monitoring signal or ground monitoring signal, thus realizing the resampling of the vehicle-mounted monitoring signal or ground monitoring signal.
[0037] In this embodiment, resampling solves the problem of inconsistent data lengths caused by different sampling frequencies between vehicle-mounted and ground-based monitoring systems; wherein... Figure 6 The data shows the alignment of the lengths of the onboard and ground monitoring signals for the track circuit in section 1710BG.
[0038] S6. Obtain the onboard monitoring signal and ground monitoring signal of any track section. After executing steps S1-S5, input them into a bidirectional LSTM neural network for training to generate a fault trend early warning model for poor routing.
[0039] In this embodiment, the structure and connection relationships of the bidirectional LSTM neural network are as follows: Figure 7 As shown, the model includes a first-layer bidirectional LSTM neural network, a second-layer bidirectional LSTM neural network, a self-attention mechanism layer, a fully connected layer, and a conditional random field layer. The first and second layers each include both forward and backward LSTM layers, each composed of multiple LSTM cell units. The self-attention mechanism layer enables the bidirectional LSTM neural network model to capture the relationship between signals at any two locations. The fully connected layer acts as a classifier, outputting the probability of the data's category based on the learned high-dimensional features. The conditional random field layer models the dependencies between labels using global conditional probabilities, optimizing the model's recognition results. The training process of the bidirectional LSTM neural network is as follows: Obtain vehicle-mounted monitoring signals and ground monitoring signals from any tunnel section, and execute steps S1-S5 to generate resampled vehicle-mounted monitoring signals and ground monitoring signals.
[0040] Acquire the induced voltage data of the resampled vehicle-mounted monitoring signal, and simultaneously acquire the power output voltage, power output current, sending-end cable side voltage, sending-end cable side current, receiving-end cable side main rail voltage, and receiving-inlet main rail voltage of the resampled ground monitoring signal to form a dataset.
[0041] The dataset is divided into training set, validation set and test set according to a set ratio. The training set has labeled data, that is, each data point in the training set (including induced voltage data point, power output voltage data point, power output current data point, sending end cable side voltage data point, sending end cable side current data point, receiving end cable side main rail voltage data point and receiving inlet main rail voltage data point) is labeled. 0 indicates that the branch is normal, 1 indicates that the branch is faulty warning and 2 indicates that the branch is faulty.
[0042] The training set is input into the bidirectional LSTM neural network for training, resulting in the trained bidirectional LSTM neural network. The process of inputting the training set into the bidirectional LSTM neural network for training is as follows: Step 1: Given the input sequence , , , , These represent the 1st, 2nd, and 3rd elements of the input sequence, respectively. The, the One element, Indicates the total length of the data.
[0043] Step 2: In the forward computation of the bidirectional LSTM neural network, the input sequence... This will be passed to a standard LSTM neural network cell, which processes the input sequence by maintaining a hidden state and a cell state. Figure 8 For time The calculation process for the forward cell unit is as follows:
[0044]
[0045]
[0046]
[0047]
[0048]
[0049] in, , The first Real-time cell unit state and output; , The first Output of cell unit status at any given time; For the first Input data at any given time; , These are the forget gate and the input gate, used to calculate the cell unit at the previous time step. The weights of each feature are determined by... , After the Sigmoid activation function ( ) Calculated, , Both are vectors with values between [0, 1]. A vector element equal to 1 means that the corresponding feature is completely retained in the calculation, while an element equal to 0 means that the corresponding feature is completely discarded in the calculation. Depend on and This is an intermediate result obtained through the hyperbolic tangent activation function (tanh). This is the intermediate calculation result of the output gate; , , , , , , , These are all parameters to be trained on the model.
[0050] Step 3: Reverse the input sequence to obtain... , Indicates the first Each element, calculated in the same steps as in Step 2, yields the time. Reverse cell unit output .
[0051] Step 4: Hidden states generated by the forward cell units Hidden states generated by reverse cell units At every moment The fusion is achieved through splicing, and the calculation is as follows:
[0052] in, For the first The two-way hidden state at any given moment.
[0053] Step 5: The output is the hidden state sequence after forward and backward computation. , , , Representing time 1, time 2, and time 3 respectively. The two-way hidden state at any given moment.
[0054] Step 6: Contain the hidden state sequence The inputs are sequentially processed through a self-attention mechanism layer, a fully connected layer, and a conditional random field layer. Through feature extraction and classification, the trained bidirectional LSTM neural network is finally obtained.
[0055] Simultaneously, the performance of the trained bidirectional LSTM neural network is evaluated using a validation set to obtain the optimal bidirectional LSTM neural network.
[0056] In this embodiment, after training the model in the above steps, the model performance is evaluated on the validation set. By evaluating the performance of different network structures or hyperparameter models on the validation set, the model with the best performance is selected as the optimal bidirectional LSTM neural network.
[0057] The test set is input into the optimal bidirectional LSTM neural network to evaluate the ability to predict the trend of faulty branch circuits. Finally, a trained bidirectional LSTM neural network is obtained and used as a model for predicting the trend of faulty branch circuits.
[0058] S7. Obtain the onboard monitoring signal and ground monitoring signal of the track section to be tested. After executing steps S1-S5, input the shunt failure trend early warning model to predict the local shunt status. At the same time, introduce the peak method to obtain the compensation capacitor information of the track section to be tested. Combine the local shunt status prediction results to locate the shunt failure trend location.
[0059] Specifically, step S7 includes: After acquiring the onboard monitoring signal and ground monitoring signal of the track section to be tested, and executing steps S1-S5, resampled onboard monitoring signal and ground monitoring signal of the track section to be tested are generated.
[0060] The resampled onboard monitoring signal and ground monitoring signal of the track section to be tested are input into the fault trend early warning model for poor shunting to predict the local location shunting status, and generate the shunting failure result of each data point of the track section to be tested, which is a warning of whether the shunting is normal, poor, or poor.
[0061] The peak value method is used to calculate the peak value of the compensation capacitor induced pulse data in the resampled onboard monitoring signal of the track section under test, and the ground compensation capacitor location information is obtained, specifically: Set size to The sliding window searches from the starting point of the compensation capacitor induced pulse signal in the resampled onboard monitoring signal of the track section under test, and obtains the local peak value of the sliding window. , Indicates the first A local peak.
[0062] In this embodiment, the on-board monitoring signal includes the main rail signal and the compensation capacitor induction pulse signal, which correspond one-to-one. The starting point of the main rail signal is the starting point of the compensation capacitor induction pulse signal.
[0063] judge Is it 1 or a local peak value of the current sliding window? If the distance is greater than 50, then retain the local peak value. Move the sliding window to the right to find the next local peak; otherwise, retain the local peak. With local peak The maximum value in the range is obtained, and the sliding window is moved to the right to find the next local peak value. Finally, the compensation capacitance and position of the track segment to be tested are obtained, i.e.:
[0064] in, This indicates the number of compensation capacitors in the track section under test. , , These represent the 1st, 2nd, and 3rd segments of the track segment to be measured, respectively. One compensation capacitor.
[0065] Based on the shunt failure results of each data point in the track section under test and the location of the compensation capacitor in the track section under test, the location of the shunt failure trend can be determined.
[0066] In this embodiment, the pulse signal sensed by the integrated inspection vehicle when passing directly above the rail compensation capacitor is the local maximum value (which reflects the train's running position), thus obtaining the correspondence between the compensation capacitor position and the data number. For example, the 59th point of the compensation capacitor's sensed pulse signal is the first local peak, which is identified by C1; the 159th point is the second local peak, which is identified by C2, and so on, finally obtaining the correspondence between C1, C2, ... and the data number; then, the horizontal axis (data number) is replaced with the compensation capacitor number to obtain the fault trend position of the shunt defect; for example, the experimental results after replacement are as follows: the model identifies a shunt defect fault trend from C2 to C9 (corresponding to the train running from C2 to C9), and subsequently, the shunt condition of this part is investigated based on the actual installation positions of C2 and C9 on site.
[0067] To verify the effectiveness of the ZPW-2000 track circuit shunt failure trend early warning algorithm proposed in this invention, the following experiments were conducted: This invention uses onboard monitoring data from the Laiyang South Electrification Section of the Weirong Expressway under the Jinan Railway Bureau from May to July 2024, and monitoring data from the Centralized Signaling Monitoring (CSM) system for verification; among which, Figure 4 , Figure 5 and Figure 6 These represent the onboard monitoring induced voltage and the main rail voltage at the ground receiving inlet for section 1710BG before preprocessing, after time alignment, and after fitting resampling operations; from... Figure 4 It can be seen that before the time alignment operation, the vehicle-mounted monitoring signal was continuous and unsegmented, and there was also a time inconsistency with the ground monitoring signal. After the S1-S4 operations of this invention, time-aligned vehicle-to-ground data can be obtained, such as... Figure 5 As shown; due to the inconsistency between the sampling frequency of the vehicle-mounted monitoring (one point every 2.5m) and the sampling frequency of the ground monitoring signal (4Hz), the amount of data collected in the same time period differs. To address this, this invention fits the data curve using cubic spline interpolation, and then samples at twice the frequency of the ground monitoring signal to ensure consistent vehicle-to-ground data length. The fitted and resampled vehicle-to-ground data curve is shown below. Figure 6This shows that the resampling curve is basically the same as the original curve, and the vehicle-to-ground data length is the same.
[0068] This invention utilizes the local peak characteristics of the compensation capacitor induction in onboard monitoring signals. By traversing the induction capacitor data, it identifies local maxima, thereby adaptively determining the location and quantity of compensation capacitors in the track circuit section. This method provides crucial location information for subsequent shunt failure trend early warning and location; experimental results... Figure 9 As shown, the identification results of the compensation capacitor location information of the ground monitoring signal of the track circuit in section 1710BG are presented.
[0069] Therefore, after preprocessing steps S1-S5, a total of 456 data entries were obtained. The dataset was divided into training, validation, and test sets in a 4:1:1 ratio for 5-fold cross-validation. The training set was used to optimize network parameters, and the network structure, learning rate, and other hyperparameters were adjusted based on the performance on the validation set. The optimal network structure was a bidirectional LSTM with 128 hidden layers, using the Adam optimizer, and a learning rate and weight decay rate of 0.00056 and 0.001, respectively. To better illustrate the effectiveness of the fault trend warning model for poor shunt circuits, two track circuit sections (section 1430AG and section 1568BG) were randomly selected and input into the trained model for testing. The experimental results are as follows: Figure 10 and Figure 11 As shown; from the perspective of the main rail voltage at the receiving input of the ground monitoring signal, in Figure 10 During the process, the voltage at the rail outlet drops rapidly after the train enters the circuit, and the residual voltage of the shunt stabilizes at around 3mV. Therefore, there is almost no local shunt failure and no tendency for shunt failure to occur. Figure 11 The first half of the section has a large residual voltage in the branch circuit, and there is a local branch circuit failure, indicating a trend of branch circuit failure. The phenomenon can be located at C2-C9 by the compensation capacitor signal, and a branch circuit failure trend warning information should be given. It can be seen that the branch circuit failure trend warning model proposed in this invention can accurately detect the occurrence of local branch circuit failure, give a branch circuit failure trend warning information, and locate its location.
[0070] With network structure and other parameters fixed, the performance of the model on the test set is evaluated. The evaluation metrics used are precision, recall, and macro-F1, calculated as follows:
[0071]
[0072]
[0073] in, , and They represent the first The number of positive samples correctly classified, negative samples correctly classified, and negative samples misclassified in each class of samples. The total number of sample classes; obtain the 5 macro-F1 scores for the test set, as follows: Figure 12 As shown, the average macro-F1 score is 0.9682, and the average confusion matrix of the 5 results is as follows. Figure 13 As shown.
[0074] Based on the comprehensive experimental results, the fault trend early warning model for shunt circuit defects proposed in this invention can accurately identify local shunt circuit defects, provide fault trend early warning information for shunt circuit defects, and locate the location where they occur. Therefore, the fault trend early warning algorithm for shunt circuit defects in the ZPW-2000 type track circuit proposed in this invention, which is a method based on vehicle-ground monitoring joint analysis, can be used as a new technology or method for preventing shunt circuit defects in railway field.
[0075] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.
[0076] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.
Claims
1. A fault trend early warning algorithm for faulty circuit shunts in ZPW-2000 type track circuits, characterized in that, Includes the following steps: S1. Acquire vehicle-mounted monitoring signals and ground monitoring signals; S2. Based on the carrier frequency crossover configuration characteristics of adjacent track circuits in the interval, the on-board monitoring signal is segmented to generate segmented on-board monitoring signals. S3. Move the sliding window of the segmented onboard monitoring signals to generate the main track signal; S4. Acquire the ground monitoring signal from the main track signal from the start time to the set moving time threshold, move the sliding window, and generate the time-aligned vehicle monitoring signal and ground monitoring signal. S5. Perform cubic spline interpolation and resampling on the time-aligned vehicle-mounted monitoring signal and ground monitoring signal respectively to generate resampled vehicle-mounted monitoring signal and ground monitoring signal. S6. Obtain the onboard monitoring signal and ground monitoring signal of any track section. After executing steps S1-S5, input them into a bidirectional LSTM neural network for training to generate a fault trend early warning model for poor routing. S7. Obtain the onboard monitoring signal and ground monitoring signal of the track section to be tested. After executing steps S1-S5, input the shunt failure trend early warning model to predict the local shunt status. At the same time, introduce the peak method to obtain the compensation capacitor information of the track section to be tested. Combine the local shunt status prediction results to locate the shunt failure trend location.
2. The ZPW-2000 track circuit shunt failure trend early warning algorithm according to claim 1, characterized in that, The on-board monitoring signals are monitoring information from the track inspection vehicle or locomotive signal information, including time, signal, carrier frequency, induced voltage, and induced pulse data from the compensation capacitor.
3. The ZPW-2000 track circuit shunt failure trend early warning algorithm according to claim 1, characterized in that, Ground monitoring signals are track circuit monitoring information, including power output voltage, power output current, voltage on the sending cable side, current on the sending cable side, main rail voltage on the receiving cable side, and main rail voltage data at the receiving entrance.
4. The ZPW-2000 track circuit shunt failure trend early warning algorithm according to claim 1, characterized in that, Step S2 specifically includes: Starting from the starting point of the vehicle-mounted monitoring signal, record the carrier frequency along the route and find the points of change of the carrier frequency; The first change point of the carrier frequency is taken as the starting point of the current track segment and the ending point of the previous track segment, and the second change point of the carrier frequency is taken as the ending point of the current track segment and the starting point of the next track segment, thus realizing the segmentation of the vehicle monitoring signal.
5. The ZPW-2000 track circuit shunt failure trend early warning algorithm according to claim 1, characterized in that, Step S3 specifically includes: Set size to The sliding window, based on the induced voltage data in the segmented vehicle monitoring signal, calculates adjacent values sequentially from the receiver to the transmitter. Construct the first variance array based on the variance between the points. ; For the first variance array A traversal search is performed, and the first point where the variance changes drastically is taken as the starting point of the induced voltage data. At the same time, the second point where the variance changes drastically is taken as the ending point of the induced voltage data. Finally, the induced voltage data from the starting point to the ending point is taken as the main track signal.
6. The ZPW-2000 track circuit shunt failure trend early warning algorithm according to claim 1, characterized in that, Step S4 specifically includes: Acquire ground monitoring signals from the main orbit signal from the start time to a set movement time threshold; Set size to The sliding window, based on the main rail voltage data received from the ground monitoring signal reception entrance, calculates adjacent values sequentially from the starting point to the ending point. Construct a second variance array based on the variance between points. ; For the second variance array Perform a traversal search, take the first point where the variance changes drastically as the starting point of the train entering the track section, and take the second point where the variance changes drastically as the ending point of the train leaving the track section, and finally obtain the time of the starting point and the ending point. Ground monitoring data is extracted from the time period from the start point to the end point to achieve time alignment between vehicle-mounted monitoring signals and ground monitoring signals.
7. The ZPW-2000 track circuit shunt failure trend early warning algorithm according to claim 1, characterized in that, Step S5 specifically includes: A set of discrete data points is used to represent time-aligned vehicle-mounted or ground-based monitoring signals, i.e.: in, , , , , They represent the 1st, 2nd, and 3rd respectively. The, the The, the The time for each discrete data point , , , , They represent the 1st, 2nd, and 3rd respectively. The, the The, the The values of discrete data points; For each pair of adjacent discrete data points and discrete data points Using cubic interpolation polynomials Describe the adjacent interval The cubic interpolation, i.e.: in, Indicates adjacent intervals The cubic interpolation polynomial, , , , Each represents an unknown coefficient to be solved. Indicates the time of the original data; Set a cubic interpolation polynomial The second derivative at the boundary is zero, and for every pair of adjacent discrete data points... and discrete data points cubic interpolation polynomial All satisfy the condition that the cubic interpolation polynomials of adjacent intervals are equal at the data points, the first derivative, and the second derivative, so as to generate a system of linear equations containing unknown coefficients. The Gaussian elimination method is used to solve the system of linear equations containing unknown coefficients, ultimately generating the cubic interpolation polynomial for the solution. ; By splicing together the cubic interpolation polynomials of each adjacent interval, the fitting curve function of the discrete data points is obtained, that is, the fitted vehicle monitoring signal or ground monitoring signal is obtained, realizing the cubic spline interpolation of the time-aligned vehicle monitoring signal or ground monitoring signal. The fitted vehicle-mounted monitoring signal or ground monitoring signal is sampled at equal intervals, and the number of sampling points is twice the length of the ground monitoring signal before fitting, to generate a resampled vehicle-mounted monitoring signal or ground monitoring signal, thus realizing the resampling of the vehicle-mounted monitoring signal or ground monitoring signal.
8. The ZPW-2000 track circuit shunt failure trend early warning algorithm according to claim 1, characterized in that, A bidirectional LSTM neural network consists of a first bidirectional LSTM neural network layer, a second bidirectional LSTM neural network layer, a self-attention mechanism layer, a fully connected layer, and a conditional random field layer.
9. The ZPW-2000 track circuit shunt failure trend early warning algorithm according to claim 1, characterized in that, Step S6 specifically includes: Obtain vehicle-mounted monitoring signals and ground monitoring signals from any tunnel section, and execute steps S1-S5 to generate resampled vehicle-mounted monitoring signals and ground monitoring signals; Acquire the induced voltage data of the resampled vehicle-mounted monitoring signal, and simultaneously acquire the power output voltage, power output current, sending-end cable side voltage, sending-end cable side current, receiving-end cable side main rail voltage and receiving-inlet main rail voltage of the resampled ground monitoring signal to form a dataset. The dataset is divided into training set, validation set and test set according to a set ratio. The training set has labeled data, that is, each data point in the training set has a label. 0 indicates normal routing, 1 indicates bad routing warning, and 2 indicates bad routing. The training set is input into the bidirectional LSTM neural network for training, and the performance of the trained bidirectional LSTM neural network is evaluated using the validation set to obtain the optimal bidirectional LSTM neural network. The test set is input into the optimal bidirectional LSTM neural network to evaluate the ability to predict the trend of faulty branch circuits. Finally, a trained bidirectional LSTM neural network is obtained and used as a model for predicting the trend of faulty branch circuits.
10. The ZPW-2000 track circuit shunt failure trend early warning algorithm according to claim 1, characterized in that, Step S7 specifically includes: After obtaining the onboard monitoring signal and ground monitoring signal of the track section to be tested, and executing steps S1-S5, the resampled onboard monitoring signal and ground monitoring signal of the track section to be tested are generated. The resampled onboard monitoring signal and ground monitoring signal of the track section to be tested are input into the fault trend early warning model for poor shunting to predict the local location shunting status, and generate the shunting failure result of each data point of the track section to be tested, which is a warning of normal shunting, poor shunting or poor shunting. The peak value method is used to calculate the peak value of the compensation capacitor induced pulse data in the resampled onboard monitoring signal of the track section under test, and the ground compensation capacitor location information is obtained, specifically: Set size to The sliding window searches from the starting point of the compensation capacitor induced pulse signal in the resampled onboard monitoring signal of the track section under test, and obtains the local peak value of the sliding window. , Indicates the first A local peak; judge Is it 1 or a local peak value of the current sliding window? If the distance is greater than 50, then retain the local peak value. Move the sliding window to the right to find the next local peak; otherwise, retain the local peak. With local peak The maximum value in the range is obtained, and the sliding window is moved to the right to find the next local peak value, finally obtaining the compensation capacitance of the track segment to be tested, i.e.: in, This indicates the number of compensation capacitors in the track section under test. , , These represent the 1st, 2nd, and 3rd segments of the track segment to be measured, respectively. One compensation capacitor; Based on the shunt failure results of each data point in the track section under test and the location of the compensation capacitor in the track section under test, the location of the shunt failure trend can be determined.