An automated cleaning method for track monitoring data based on deep learning

Abstract

The present invention discloses an automatic cleaning method for track monitoring data based on deep learning, which belongs to the field of railway monitoring data cleaning. This method uses the SL-MIL algorithm to eliminate irregular factors of the track profile by establishing a prediction model between the track profile and the track monitoring data. Considering the impact of multi-track driving on the actual monitoring data, the LA-VD algorithm is introduced to realize the automatic identification of train monitoring data with different axle weights and driving speeds, and to judge in real time the interference of the remaining tracks to the required monitoring tracks when there are trains passing, and to automatically clean the interference data. The algorithm significantly improves the calculation efficiency and reduces the calculation time, thereby achieving more accurate and effective data cleaning. It has the advantages of high adaptability, high precision, high reliability, etc., and automatically adjusts the frequency of the modulation function according to different monitoring data conditions, and adapts to the extraction of frequency domain data affected by different conditions.

Description

Technical Field

[0001] The present invention relates to the field of railway monitoring data cleaning, and in particular to an automatic cleaning method for track monitoring data based on deep learning. Background Art

[0002] With the rapid development of the economy, railway transportation and rail transportation play a very important role in the field of transportation. Especially with the continuous improvement of the integration, standardization, integration and complexity of train operations, the safety and data analysis in rail transportation operations are becoming more and more important, and the stability of railway operations is becoming increasingly demanding. In order to obtain accurate real-time data, various sensor devices are now mainly used to collect real-time data during rail transportation operation, and the health status of the rail transportation is evaluated and warned based on the collected data, which greatly reduces the unexpected situations that occur during the operation of rail transportation. By processing the data obtained from monitoring, the problem area can be quickly known, which is conducive to efficient equipment maintenance.

[0003] Data cleaning plays an important role in the field of rail transit. In the actual application of rail transit structure health monitoring, factors such as failure of arranged sensors, interference in data transmission, and other interference in adjacent locations often occur, resulting in monitoring information with implicit erroneous information in the monitored data. In addition, due to the presence of abnormal values ​​in the monitoring information, erroneous recognition results are caused, resulting in erroneous assessments of the health status of the rail transit structure, resulting in a large amount of manpower and material resources being wasted for inspection and maintenance, which seriously affects the normal operation of the rail transit structure.

[0004] At the same time, in actual train tracks, the rail profiles at different positions are different. The wear of the rails at different positions affects the accuracy of the track monitoring data at that position to a certain extent. Considering that there are multiple tracks with vehicles coming at the same time during track driving in reality, the passing trains on the remaining tracks will affect the track area to be monitored, causing errors in the monitoring data. Therefore, it is very important to propose an automatic cleaning method for track monitoring data. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides an automatic cleaning method for track monitoring data based on deep learning.

[0006] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0007] A deep learning-based automatic cleaning method for track monitoring data comprises the following steps:

[0008] S1. Collect and classify the monitoring data of the tracks in the monitoring area to obtain a data set;

[0009] S2. Establish and train a prediction model between track profile and track monitoring data, obtain a data set of abnormal track monitoring data, and perform preliminary cleaning;

[0010] S3. Establish an automatic identification and cleaning model for multi-track traffic interference and track monitoring data, and automatically clean the interference data;

[0011] S4. Real-time calculation of the wheel load reduction rate and derailment coefficient when the train passes by to achieve real-time evaluation of rail safety.

[0012] Furthermore, the S2 specifically includes the following steps:

[0013] S21, using a data set formed by monitoring data of a standard track profile as a training sample to train a prediction model constructed;

[0014] S22, using a probability model algorithm to obtain a weight distribution probability of the monitoring data after cleaning under the condition of inputting a standard track profile, a track shape and a size parameter;

[0015] S23. Complete the track monitoring data under the condition of track profile impact data cleaning according to the weight distribution probability obtained in S22.

[0016] Furthermore, the prediction model in S21 is expressed as:

[0017] ;

[0018] ;

[0019]

[0020] ;

[0021] In the formula, is the output data of the model, which refers to the monitoring data of the track; for dimensional implicit matrix, is the number of hidden layer nodes, is the number of track monitoring data; Indicates The weights of the hidden layer nodes; For the The bias of hidden layer nodes; is the starting function; is the outline of the track; for dimensional monitoring data prediction error vector; is the weight of the output layer; It represents the distribution probability of the weight ratio of the monitoring data after cleaning outputted under the premise of inputting the track profile and the track shape and size parameters; is the gamma function; is the shape parameter of the probability distribution of the orbit; is the scale parameter of the probability distribution of the orbit; is the shape parameter of the track; is the size parameter of the track; Indicates that the regularization parameter Extract the diagonal elements of the vector matrix and create a diagonal matrix; Indicates that in the regular A hyperparameter vector matrix is ​​added to the matrix.

[0022] Furthermore, the S3 specifically includes the following steps:

[0023] S31, numbering the tracks of different tracks respectively;

[0024] S32, counting and classifying the previous track monitoring data, obtaining the impact data of trains with different axle weights and different running speeds on the adjacent tracks when running on any track, and forming an impact data set with the impact data;

[0025] S33. Automatically determine the axle weight and running speed of a train passing through the track corresponding to any monitoring data through the LA-VD algorithm, and obtain the impact on the monitoring data when trains pass through other tracks.

[0026] Furthermore, the S33 specifically includes the following steps:

[0027] S331. The bandwidth of the data on the influence of trains passing through different tracks on the monitoring data is estimated. At the same time, the Lagrangian operator is used to transform the original problem into an unconstrained variational problem, and the influence of trains passing through different tracks on the monitoring data is obtained, which is expressed as:

[0028] ;

[0029] ;

[0030] in, The total number of numbers indicating the impact of a train passing through any track on the monitoring data of another track; Indicates the impact data set on the monitoring data when trains pass through different tracks; It is expressed as the center frequency of the affected data when a train passes through each track; To monitor the data impulse function at time The value at time; To provide monitoring data after cleaning the impact of track profile; Indicates that it is subject to conditions; is the ladder square norm of the modulation signal; represents the partial derivative with respect to time; Indicated in In this case, at time The value at time; is the instantaneous frequency under the setting condition; For time; The phase of the monitoring data; is the square of the norm, representing the monitoring data signal; means summing the squares of the norms in all cases; is a regularization parameter used to control the smoothness of modal decomposition under different conditions; Indicates that at time Monitoring data after cleaning the track profile; Lagrange multipliers for solving the unconstrained variational problem for orbit monitoring data; is the inner product between two data;

[0031] S332. According to different classification situations of the monitoring data, the size of the square norm of the modulated signal ladder is judged. If it is not within the tolerance value, each parameter in the formula is automatically adjusted. The automatic adjustment formula is:

[0032] ;

[0033] ;

[0034] ;

[0035] ;

[0036] ;

[0037] in, and They represent the upper and lower limits of the tolerance of the square norm of the modulated signal ladder, To monitor the phase of the data, is a regularization parameter used to control the smoothness of modal decomposition under different conditions. For time, To monitor the data impulse function at time The value of To solve the Lagrange multiplier of the unconstrained variational problem of orbit monitoring data, Indicates the automatic adjustment value of the monitoring data phase. represents the adjustment value of the regularization parameter under different circumstances, Indicates the automatic adjustment value of time, Indicates that at time The adjustment value of the data impulse function is monitored at the same time. represents the automatic adjustment value of the Lagrange multiplier for solving the unconstrained variational problem of track monitoring data, It is an automatic adjustment function.

[0038] Furthermore, the S4 specifically includes the following steps:

[0039] S41, using the track displacement monitoring data after cleaning to calculate the wheel load reduction rate and derailment coefficient of the track in real time;

[0040] S42, setting thresholds for the track wheel load reduction rate and derailment coefficient under different conditions, and issuing an alarm when the value calculated in real time by S41 exceeds the set threshold.

[0041] Furthermore, the specific method of using the track displacement monitoring data after cleaning to calculate the wheel load reduction rate and derailment coefficient of the track in real time in S41 is:

[0042] ;

[0043] ;

[0044] ;

[0045] ;

[0046] ;

[0047] In the formula, is the maximum train stress on the track when a train passes. E is the elastic modulus of the rail, is the section moment of inertia of the rail; The track displacement monitoring data after cleaning. is the train stress at the end of the track with the larger train stress, is the train stress at the end of the track where the train stress is smaller, is the wheel weight reduction of the wheel on the load-reducing side, is the average static wheel weight of the wheels on the unloaded and loaded sides, is the lateral force acting on the wheel, is the friction coefficient at the wheel rim, is the maximum wheel rim contact angle.

[0048] The present invention has the following beneficial effects:

[0049] The present invention takes into account the influence of track profile on actual monitoring data, establishes a prediction model between track profile and track monitoring data, and uses SL-MIL algorithm to eliminate irregular factors of track profile. Compared with other neural network models, the model has a relatively simple structure, can more quickly calculate the weight distribution probability of data points and obtain the monitoring data after cleaning, which greatly improves the efficiency of data cleaning; the present invention takes into account the influence of multi-track driving on actual monitoring data, introduces LA-VD algorithm, realizes automatic identification of train monitoring data with different axle weights and driving speeds, judges in real time the interference of other tracks to the required monitoring track when there are trains passing, and automatically cleans the interference data, the algorithm significantly improves the calculation efficiency, reduces the calculation time, and thus realizes more accurate and effective data cleaning; the present invention processes the cleaned data in real time, and realizes real-time evaluation of rail safety by calculating the wheel weight reduction rate and derailment coefficient when the train passes in real time, thereby improving the reliability and comprehensiveness of track safety evaluation; the present invention has the advantages of high adaptability, high precision, high reliability, etc., automatically adjusts the frequency of the modulation function according to different monitoring data conditions, and adapts to the extraction of frequency domain data affected by different conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of the automatic cleaning method of track monitoring data based on deep learning in the present invention;

[0051] Figure 2 It is a schematic diagram of the result of data cleaning for monitoring data in the present invention. DETAILED DESCRIPTION

[0052] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0053] A deep learning-based automated cleaning method for track monitoring data, such as Figure 1 As shown, the following steps are included:

[0054] S1. Collect and classify the monitoring data of the tracks in the monitoring area to obtain a data set;

[0055] S2. Establish and train a prediction model between track profile and track monitoring data, obtain a data set of abnormal track monitoring data, and perform preliminary cleaning;

[0056] In this embodiment, the following steps are specifically included:

[0057] S21, using a data set formed by monitoring data of a standard track profile as a training sample to train a prediction model constructed;

[0058] Input data set formed by monitoring data of standard track profile As training samples, we train the SL-MIL model.

[0059] S22, using a probability model algorithm to obtain a weight distribution probability of the monitoring data after cleaning under the condition of inputting a standard track profile, a track shape and a size parameter;

[0060] S23. Complete the track monitoring data under the condition of track profile impact data cleaning according to the weight distribution probability obtained in S22.

[0061] In this embodiment, the SL-MIL algorithm model is as follows. Compared with other neural network models, the model has a relatively simple structure. By introducing a priori of sparsity, irrelevant hidden layer nodes are automatically excluded to optimize the model structure, and the uncertainty of the posterior is effectively quantified, so that the weight distribution probability of the data point can be calculated more quickly and the monitoring data after cleaning can be obtained:

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] in, is the output data of the model, which refers to the monitoring data of the track; for dimensional implicit matrix, is the number of hidden layer nodes, is the number of track monitoring data; Indicates The weight of the hidden layer node is a D-dimensional vector randomly generated in the uniform distribution [-1, 1], where D is also the number of monitoring sensors; For the The bias of the hidden layer nodes is a scalar randomly generated from the uniform distribution [-1, 1]; As the starting function, ; is the input data of the model, which refers to the profile of the track; for dimensional monitoring data prediction error vector; is the weight of the output layer; It represents the distribution probability of the weight ratio of the monitoring data after cleaning outputted under the premise of inputting the track profile and the track shape and size parameters; is the gamma function; is the shape parameter of the probability distribution of the orbit; is the scale parameter of the probability distribution of the orbit; is the shape parameter of the orbit, ; is the size parameter of the track, ; Indicates that the regularization parameter Extract diagonal elements of vector matrix and create diagonal matrix; Indicates that in the regular A hyperparameter vector matrix is ​​added to the matrix to improve the accuracy of the prediction error.

[0067] S3. Establish an automatic identification and cleaning model for multi-track traffic interference and track monitoring data, and automatically clean the interference data;

[0068] In this embodiment, the following steps are specifically included:

[0069] S31, numbering the tracks of different tracks respectively, from the inside to the outside, 0, 1, 2, and so on;

[0070] S32. Count and classify the previous track monitoring data to obtain the impact data of trains with different axle weights and different running speeds on the adjacent tracks when running on any track, and use the impact data to form an impact data set. ;

[0071] S33. Automatically determine the axle weight and running speed of a train passing through the track corresponding to any monitoring data through the LA-VD algorithm, and obtain the impact on the monitoring data when trains pass through other tracks.

[0072] First, the square norm of the modulated signal ladder is calculated, and the bandwidth of the data on the impact of trains passing through different tracks on the monitoring data is estimated. At the same time, the Lagrangian operator is used to transform the original problem into an unconstrained variational problem, and the impact of trains passing through different tracks on the monitoring data is obtained:

[0073] ;

[0074] ;

[0075] in, The total number of numbers indicating the impact of a train passing through any track on the monitoring data of another track; Indicates the impact data set on the monitoring data when trains pass through different tracks; It is expressed as the center frequency of the affected data when a train passes through each track; To monitor the data impulse function at time The value at time; To provide monitoring data after cleaning the impact of track profile; Indicates that it is subject to conditions; is the ladder square norm of the modulation signal; represents the partial derivative with respect to time; Indicates that in the kth case at time The value at time; is the instantaneous frequency under the setting condition; For time; The phase of the monitoring data; is the square of the norm, representing the monitoring data signal; means summing the squares of the norms in all cases; is a regularization parameter used to control the smoothness of modal decomposition under different conditions; Indicates that at time Monitoring data after cleaning the track profile; Lagrange multipliers for solving the unconstrained variational problem for orbit monitoring data; is the inner product between two data;

[0076] At the same time, according to the different classifications of the monitoring data, the size of the square norm of the modulated signal ladder is judged. If it is not within the tolerance value, the parameters in the formula are automatically adjusted. The automatic adjustment formula is as follows:

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] ;

[0082] in, and They represent the upper and lower limits of the tolerance of the square norm of the modulated signal ladder, To monitor the phase of the data, is a regularization parameter used to control the smoothness of modal decomposition under different conditions. For time, To monitor the data impulse function at time The value of To solve the Lagrange multiplier of the unconstrained variational problem of orbit monitoring data, Indicates the automatic adjustment value of the monitoring data phase. represents the adjustment value of the regularization parameter under different circumstances, Indicates the automatic adjustment value of time, Indicates that at time The adjustment value of the data impulse function is monitored at the same time. represents the automatic adjustment value of the Lagrange multiplier for solving the unconstrained variational problem of track monitoring data, To automate the adjustment function

[0083] It should be noted that when the square norm of the modulated signal ladder is lower than the lower limit of its tolerance value, its parameters will be automatically adjusted at this time, and the monitoring data phase, regularization parameter and time will be adjusted first to ensure the smoothness of the monitoring data under different conditions; when the square norm of the modulated signal ladder is higher than the lower limit of its tolerance value, the impulse function value of the measured data and the Lagrange multiplier will be adjusted to ensure the accuracy of the monitoring data when cleaning. At the same time, when the parameters are automatically adjusted, the step size of each adjustment iteration is about one tenth of the initial parameters, and then it is calculated again The size of the track is calculated and compared again until the constraint condition is met; the influence data of the remaining tracks are cleaned in real time in the monitoring data to remove the influence of the remaining tracks on the monitoring data when the trains are running, and the cleaning of the monitoring data is completed.

[0084] S4. Real-time calculation of the wheel load reduction rate and derailment coefficient when the train passes by to achieve real-time evaluation of rail safety.

[0085] In this embodiment, the track displacement monitoring data after cleaning is obtained, such as Figure 2 As shown, the horizontal axis is the time in seconds (s), and the vertical axis is the displacement in millimeters (mm). Based on the obtained monitoring data, the wheel load reduction rate and derailment coefficient of the track are calculated and stored in real time;

[0086] Based on the obtained track displacement monitoring data, the wheel load reduction rate and derailment coefficient when the train passes are calculated in real time:

[0087] ;

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] In the formula, is the maximum train stress on the track when a train passes. E is the elastic modulus of the rail, is the section moment of inertia of the rail; The track displacement monitoring data after cleaning. is the train stress at the end of the track with the larger train stress, is the train stress at the end of the track where the train stress is smaller, is the wheel weight reduction of the wheel on the load-reducing side, is the average static wheel weight of the wheels on the unloaded and loaded sides, is the lateral force acting on the wheel, is the friction coefficient at the wheel rim, is the maximum wheel rim contact angle;

[0093] According to relevant specifications, different assessment limits are set for the track wheel load reduction rate and derailment coefficient under different conditions. When an abnormality occurs, the time period in which the abnormal value occurs is recorded and an alarm is issued.

[0094] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes 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 a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0095] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0096] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0097] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0098] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for automatic cleaning of track monitoring data based on deep learning, characterized in that: The steps include: S1. Collect and classify the monitoring data of the tracks in the monitoring area to obtain a data set; S2. Establish and train a prediction model between track profile and track monitoring data, obtain a data set of abnormal track monitoring data, and perform preliminary cleaning; S3. Establish an automatic identification and cleaning model for multi-track traffic interference and track monitoring data, and automatically clean the interference data, which specifically includes the following steps: S31, numbering the tracks of different tracks respectively; S32, counting and classifying the previous track monitoring data, obtaining the impact data of trains with different axle weights and different running speeds on the adjacent tracks when running on any track, and forming an impact data set with the impact data; S33, automatically determining the axle weight and running speed of a track corresponding to any monitoring data when a train passes through through the track by using the LA-VD algorithm, and obtaining the impact of the monitoring data on the other tracks when a train passes through, specifically comprising the following steps: S331. The bandwidth of the data on the influence of trains passing through different tracks on the monitoring data is estimated. At the same time, the Lagrangian operator is used to transform the original problem into an unconstrained variational problem, and the influence of trains passing through different tracks on the monitoring data is obtained, which is expressed as: Where k represents the total number of the impacts on the monitoring data of another track when a train passes through any track; {u k } represents the impact data set on the monitoring data when trains pass through different tracks; {ω k } represents the central frequency of the monitoring data when a train passes through each track; δ(t) is the value of the monitoring data impulse function at time t; f is the monitoring data after cleaning the track profile; st represents the conditional constraints; L 2 is the ladder square norm of the modulation signal; represents the partial derivative with respect to time; u k (t) represents the value at time t in the kth case; ω t is the instantaneous frequency under the setting condition; t is the time; j is the phase of the monitoring data; is the square of the norm, representing the monitoring data signal; represents the sum of the squares of the norms in all cases; α is a regularization parameter used to control the smoothness of modal decomposition in different cases; f(t) represents the monitoring data after the track profile is cleaned at time t; λ(t) is the Lagrange multiplier for solving the unconstrained variational problem of track monitoring data; <·,·> is the inner product between two data; S332. According to different classification situations of the monitoring data, the size of the square norm of the modulated signal ladder is judged. If it is not within the tolerance value, each parameter in the formula is automatically adjusted. The automatic adjustment formula is: L 2 =H{j,α,t,δ(t),λ(t)}; in, and They represent the upper and lower limits of the tolerance value of the square norm of the modulated signal ladder, j is the phase of the monitoring data, α is the regularization parameter used to control the smoothness of the modal decomposition under different conditions, t is time, δ(t) is the value of the monitoring data impulse function at time t, λ(t) is the Lagrange multiplier for solving the unconstrained variational problem of track monitoring data, Δj represents the automatic adjustment value of the monitoring data phase, Δα represents the adjustment value of the regularization parameter under different conditions, Δt represents the automatic adjustment value of time, Δδ(t) represents the adjustment value of the monitoring data impulse function at time t, Δλ(t) represents the automatic adjustment value of the Lagrange multiplier for solving the unconstrained variational problem of track monitoring data, and H{} is the automatic adjustment function; S4. Real-time calculation of the wheel load reduction rate and derailment coefficient when the train passes by to achieve real-time evaluation of rail safety.

2. The method for automatic cleaning of track monitoring data based on deep learning according to claim 1, characterized in that: The S2 specifically includes the following steps: S21, using a data set formed by monitoring data of a standard track profile as a training sample to train a prediction model constructed; S22, using a probability model algorithm to obtain a weight distribution probability of the monitoring data after cleaning under the condition of inputting a standard track profile, a track shape and a size parameter; S23. Complete the track monitoring data under the condition of track profile impact data cleaning according to the weight distribution probability obtained in S22.

3. The method for automatic cleaning of track monitoring data based on deep learning according to claim 2 is characterized in that: The prediction model in S21 is expressed as: Y = G (X) ω + ε; C = diag(α) + G(X)TG(X); Where Y is the output data of the model, which refers to the track monitoring data; G(X) is the N×(M+1)-dimensional hidden matrix, M is the number of hidden layer nodes, and N is the number of track monitoring data; γ i represents the weight of the i-th hidden layer node; c i is the bias of the i-th hidden layer node; f(x) is the starting function; x i is the outline of the track; ε is the N×1-dimensional monitoring data prediction error vector; ω is the weight of the output layer; p(ω|Y,a0,b0) represents the distribution probability of the weight ratio of the cleaned monitoring data output under the premise of inputting the track profile and the track shape and size parameters; Γ(·) is the gamma function; a0 is the shape parameter of the probability distribution of the track; b0 is the scale parameter of the probability distribution of the track; a'0 is the shape parameter of the track; b'0 is the size parameter of the track; diag(α) represents extracting the diagonal elements of the vector matrix of the regularization parameter α and creating a diagonal matrix; C represents the addition of a hyperparameter vector matrix on the basis of the conventional G(X) matrix.

4. The method for automatic cleaning of track monitoring data based on deep learning according to claim 1, characterized in that: The S4 specifically includes the following steps: S41, using the track displacement monitoring data after cleaning to calculate the wheel load reduction rate and derailment coefficient of the track in real time; S42, setting thresholds for the track wheel load reduction rate and derailment coefficient under different conditions, and issuing an alarm when the value calculated in real time by S41 exceeds the set threshold.

5. The method for automatic cleaning of track monitoring data based on deep learning according to claim 4 is characterized in that: The specific method of using the track displacement monitoring data after cleaning to calculate the wheel load reduction rate and derailment coefficient of the track in real time in S41 is: In the formula, is the maximum train stress on the track when a train passes, E is the elastic modulus of the rail, I is the section inertia moment of the rail; d is the track displacement monitoring data after cleaning, P1 is the train stress at the end of the track with a larger train stress, P2 is the train stress at the end of the track with a smaller train stress, ΔP is the wheel weight reduction of the wheel on the load-reducing side, is the average static wheel weight of the wheel on the unloaded and loaded sides, Q is the lateral force acting on the wheel, μ is the friction coefficient at the wheel rim, and β is the maximum wheel rim contact angle.

Images