A track disease monitoring method and system based on improved BP neural network

Abstract

The invention discloses a track defect monitoring method and system based on an improved BP neural network. The method comprises: performing VMD decomposition on a rail vibration signal, and screening the sample entropy of an IMF signal obtained by the decomposition according to an entropy value screening feature matrix strategy to obtain a sample entropy feature matrix; extracting feature information from the rail vibration signal, and fusing the feature information with the sample entropy features of each target IMF signal to obtain fused feature information; inputting the fused feature information into an improved BP neural network, and iteratively training the improved BP neural network based on a dynamic parameter adjustment strategy to obtain a track defect monitoring model.

Description

Technical Field

[0001] The invention belongs to the technical field of rail transit detection, and in particular relates to a rail disease monitoring method and system based on an improved BP neural network. Background Art

[0002] With the increasing mileage of railway operation, under the dual influence of external environment and wheel-rail friction, track failures are becoming more and more frequent. Track structures are often subject to various irreversible damages due to environmental factors such as wheel friction, wind and sun. For example, there are many forms of fastener diseases, track plate diseases, CA mortar diseases, supporting layer or base plate diseases, and main structure diseases caused by changes in the lower roadbed structure. If these diseases are not repaired in time, they will continue to damage the track structure, affect the track operation efficiency, and shorten the service life. In order to ensure the safety, comfort and efficiency of railway operation, it is crucial to regularly inspect the track structure and carry out maintenance in a timely manner; at the same time, this is also an effective way to protect railway investment, reduce maintenance costs and extend the service life of railways.

[0003] Compared with traditional methods based on vibration signals, the BP neural network in machine learning technology can automatically seek potential connections between features and track defect categories from large amounts of data without human intervention, which can reduce the risk of human error while improving detection efficiency and accuracy, providing a new idea for track defect monitoring. Summary of the invention

[0004] The present invention provides a track disease monitoring method and system based on an improved BP neural network, which are used to solve the technical problem that the efficiency and accuracy of detection cannot be improved.

[0005] In a first aspect, the present invention provides a track disease monitoring method based on an improved BP neural network, comprising:

[0006] The rail vibration signal is obtained according to the preset signal anti-aliasing processing rule, and the expression of the signal anti-aliasing processing rule is:

[0007] ,

[0008] ,

[0009] In the formula, is the stop-band attenuation, is the Nyquist zone where the signal is located, is the sampling frequency, is the upper frequency limit of the signal, is the carrier frequency, For the integral, is the bandwidth of the collected rail vibration signal, is the 3dB bandwidth, is the lower frequency limit of the signal, is the filter order;

[0010] Perform VMD decomposition on the rail vibration signal, and screen the sample entropy of the IMF signal obtained by the decomposition according to the entropy value screening feature matrix strategy to obtain a sample entropy feature matrix, wherein the sample entropy feature matrix contains the sample entropy features of each target IMF signal;

[0011] Extracting feature information from the rail vibration signal according to a pre-built DRSN deep learning model, and fusing the feature information with sample entropy features of each target IMF signal to obtain fused feature information;

[0012] Inputting the fused feature information into an improved BP neural network, and iteratively training the improved BP neural network based on a dynamic parameter adjustment strategy to obtain a track disease monitoring model;

[0013] The acquired real-time rail vibration signal is input into the track defect monitoring model, and the track defect monitoring model outputs a monitoring result corresponding to the real-time rail vibration signal.

[0014] In a second aspect, the present invention provides a track defect monitoring system based on an improved BP neural network, comprising:

[0015] The acquisition module is configured to acquire the rail vibration signal according to a preset signal anti-aliasing processing rule, wherein the expression of the signal anti-aliasing processing rule is:

[0016] ,

[0017] ,

[0018] In the formula, is the stop-band attenuation, is the Nyquist zone where the signal is located, is the sampling frequency, is the upper frequency limit of the signal, is the carrier frequency, For the integral, is the bandwidth of the collected rail vibration signal, is the 3dB bandwidth, is the lower frequency limit of the signal, is the filter order;

[0019] A decomposition module is configured to perform VMD decomposition on the rail vibration signal, and screen the sample entropy of the IMF signal obtained by the decomposition according to the entropy value screening feature matrix strategy to obtain a sample entropy feature matrix, wherein the sample entropy feature matrix contains the sample entropy features of each target IMF signal;

[0020] A fusion module is configured to extract feature information from the rail vibration signal according to a pre-built DRSN deep learning model, and fuse the feature information with the sample entropy features of each target IMF signal to obtain fused feature information;

[0021] A training module is configured to input the fusion feature information into an improved BP neural network, and iteratively train the improved BP neural network based on a dynamic parameter adjustment strategy to obtain a track disease monitoring model;

[0022] The output module is configured to input the acquired real-time rail vibration signal into the track defect monitoring model, and the track defect monitoring model outputs a monitoring result corresponding to the real-time rail vibration signal.

[0023] In a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the track defect monitoring method based on the improved BP neural network of any embodiment of the present invention.

[0024] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor executes the steps of the track defect monitoring method based on an improved BP neural network according to any embodiment of the present invention.

[0025] The proposed track defect monitoring method and system based on the improved BP neural network can reduce the influence of ambient noise on the collected rail vibration signal, thereby making the collected signal more accurate, and the accuracy of subsequent model learning, training, and verification is also higher. The proposed entropy value screening feature matrix strategy can mine the hidden feature information in the rail signal through the fusion of VMD decomposition and deep residual network, extract the feature information contained in the rail vibration signal at multiple scales, and improve the utilization rate of the hidden information in the rail vibration signal, thereby improving the accuracy and detection efficiency of the system for rail fault diagnosis. The dynamic parameter adjustment strategy can prevent the model from being underfitted or overfitted by adjusting some hyperparameters (fitness, weights, and thresholds, etc.) in the genetic algorithm and BP neural network, and improve the efficiency, accuracy, and generalization of model learning. More importantly, the network is not only suitable for a single track defect situation, but also can achieve effective monitoring and accurate classification of multiple track defects by learning the characteristics of different track defect rail vibration signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 A flow chart of a track defect monitoring method based on an improved BP neural network provided by an embodiment of the present invention;

[0028] Figure 2 A structural block diagram of a track defect monitoring system based on an improved BP neural network provided by an embodiment of the present invention;

[0029] Figure 3 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] See also Figure 1 , which shows a flow chart of a track defect monitoring method based on an improved BP neural network of the present application.

[0032] like Figure 1 As shown in FIG. 1 , the track disease monitoring method based on the improved BP neural network specifically includes the following steps:

[0033] Step S101, obtaining a rail vibration signal according to a preset signal anti-aliasing processing rule.

[0034] In this step, the multi-channel anti-aliasing vibration signal acquisition system uses 5-channel ICP vibration sensors to collect signals and obtain rail vibration signals; the signal conditioning channel performs signal preprocessing such as filtering and amplification, bias adjustment, and single-ended to differential conversion to facilitate subsequent work; in order to prevent the mixing of different signals in the system due to insufficient sampling rate, an anti-aliasing processing rule is proposed, which is expressed as follows:

[0035] ,

[0036] ,

[0037] In the formula, is the stop-band attenuation, is the Nyquist zone where the signal is located, is the sampling frequency, is the upper frequency limit of the signal, is the carrier frequency, For the integral, is the bandwidth of the collected rail vibration signal, is the 3dB bandwidth, is the lower frequency limit of the signal, is the filter order;

[0038] Step S102, performing VMD decomposition on the rail vibration signal, and screening the sample entropy of the IMF signal obtained by the decomposition according to the entropy value screening feature matrix strategy to obtain a sample entropy feature matrix, wherein the sample entropy feature matrix contains the sample entropy features of each target IMF signal.

[0039] In this step, the rail vibration signal is subjected to a variational mode decomposition (VMD) to obtain each IMF signal, and each IMF signal is subjected to a Hilbert transform to obtain an analytical signal of each IMF signal. The expression of the single-sided spectrum of the analytical signal is:

[0040] ,

[0041] In the formula, is the impulse function, is an imaginary unit, is the kth IMF signal after decomposition, For time;

[0042] By mixing an exponential term , the center frequency of each IMF signal is estimated, and the spectrum of each IMF signal obtained after decomposition is converted to the corresponding baseband, wherein the conversion expression is:

[0043] ,

[0044] In the formula, is the center frequency corresponding to each IMF signal after decomposition;

[0045] The bandwidth of each IMF signal is estimated using the Gaussian smoothed offset signal, and the variational problem is constructed as follows:

[0046] ,

[0047] In the formula, is the original signal, It means to find the partial derivative of t. is the kth IMF signal after decomposition, is the center frequency corresponding to each IMF signal after decomposition;

[0048] To solve the variational problem, a quadratic penalty factor is introduced and the Lagrange multiplication operator , the constrained variational problem is transformed into an unconstrained variational problem. The expression of the unconstrained variational problem is:

[0049] ,

[0050] In the formula, is the Lagrangian operation equation, is the Lagrange multiplication operator, is the input signal;

[0051] After multiple iterative calculations, the minimum value of the Lagrangian expression is searched to obtain the optimal solution and the center frequency of the optimal solution, where the iterative update expression is:

[0052] ,

[0053] ,

[0054] ,

[0055] In the formula, is the frequency expression of the kth signal at the n+1th iteration, is the frequency expression of the original signal, is the frequency expression of the ith signal, is the frequency expression of the Lagrangian operator, is the center frequency of the kth signal of the n+1th iteration, is the frequency variable, is the frequency expression of the kth signal, is the time constant, is the Lagrangian operator of the n+1th iteration, is the Lagrangian operator of the nth iteration;

[0056] Given a discriminant accuracy , if the preset iteration condition is met, the iteration is stopped, where the expression of the iteration condition is:

[0057] ,

[0058] In the formula, is the kth signal of the n+1th iteration, is the kth signal of the nth iteration.

[0059] It should be noted that the IMF signal obtained by VMD decomposition is embedded into a high-dimensional space, and the vector Reconstruct the expression:

[0060] ,

[0061] In the formula, is the reconstructed signal, for The first vector of the vector, for The second vector of the vector, for The mth vector of a vector, for The first element of the kth vector of a vector, for The second element of the kth vector of a vector, for The mth element of the kth vector of a vector;

[0062] For each vector in the embedding space, find the vector All other vectors under the same condition, so that the distance between all other vectors is less than a pre-set threshold ;

[0063] Calculate the sample entropy of the IMF signal after VMD decomposition. The calculation formula is:

[0064] ,

[0065] In the formula, is the sample entropy, is the length of the time series, are the remaining vectors and vectors in the N-dimensional phase space The distance is greater than the threshold The probability of are the remaining vectors and vectors in the N-dimensional phase space The distance is less than the threshold The probability of

[0066] Taking the sample with the least IMF signal as a benchmark, the IMF signals with entropy values ​​greater than the first entropy threshold and less than the second entropy threshold are eliminated to form a sample entropy feature matrix.

[0067] Step S103: extract feature information from the rail vibration signal according to the pre-built DRSN deep learning model, and fuse the feature information with the sample entropy features of each target IMF signal to obtain fused feature information.

[0068] In this step, a residual shrinkage network (RSN) is constructed. RSN is composed of a stack of residual units. Using the idea of ​​identity mapping, a residual structure with cross-layer identity paths is introduced. The output of each residual unit can be expressed as:

[0069] ,

[0070] In the formula, For input, To expect output, is the residual function;

[0071] Construct a Squeeze and Excitation Network (SEN). SEN selectively emphasizes informative features and suppresses less useful features by using global information. The result of this module can be expressed as:

[0072] ,

[0073] In the formula, is the feature map output by the previous convolutional layer, is the product operation, is the weight vector, is the processed input vector;

[0074] Use the soft threshold function to reduce the noise of the rail vibration signal. Set a suitable positive threshold. The soft threshold function will set the input data noise characteristic value whose absolute value is lower than this threshold to zero. Its expression is:

[0075] ,

[0076] In the formula, is the output eigenvalue, For input, is the function threshold;

[0077] According to the rail vibration signal and entropy feature dimensions, a DRSN network is built, a DRSN deep learning model is trained, and feature information with the same dimension as the entropy feature is extracted;

[0078] The entropy features extracted by EMD and the feature information extracted by DRSN are concatenated to form fusion features and normalized as the input features of the improved BP neural network. The normalization process is:

[0079] ,

[0080] In the formula, is the normalized data, is the original data, is the minimum value in the data set, is the maximum value in the data set.

[0081] Step S104: input the fused feature information into the improved BP neural network, and iteratively train the improved BP neural network based on a dynamic parameter adjustment strategy to obtain a track disease monitoring model.

[0082] In this step, a BP neural network is constructed, and the dimension of the calculated sample entropy is 20, that is, the dimension of the feature vector is 20, and the number of input layers should be equal to the dimension of the feature vector, so the number of input layers is 20; there are four track structure service states corresponding to the sample vibration signal, namely normal, ballast bed hardening, rail surface collapse and rail corrugation, so the number of output layers should be 4, that is, a four-dimensional vector; the expression for determining the number of hidden layers is:

[0083] ,

[0084] ,

[0085] In the formula, is the number of input layers, is the number of hidden layers, is the number of output layers, is a tuning constant ranging from 1 to 10. The value range of z is 6 to 15, and the number of hidden layers z is 15;

[0086] The loss function of the improved BP neural network is a sample entropy loss function based on VMD decomposition to improve the learning and training efficiency and classification accuracy of the BP neural network. The expression of the sample entropy loss function is:

[0087] ,

[0088] In the formula, is the sample entropy loss function, is the number of samples in the dataset, is the probability that sample a belongs to category B, is a binary number, which is 1 if sample a belongs to category B, otherwise it is 0.

[0089] Genetic algorithm is introduced to optimize the weights and thresholds of BP neural network. The initial weights and thresholds generated by BP neural network are used as the initial population of genetic algorithm. Real number individual coding is adopted. The coding method is as follows:

[0090] ,

[0091] Where L is the encoding length, n is the number of neurons in the input layer, and h is the number of neurons in the hidden layer. is the number of neurons in the output layer;

[0092] The fitness function of the genetic algorithm is determined as:

[0093] ,

[0094] In the formula, is the fitness function, is the weight coefficient of the i-th output node, is the expected output of the i-th node in the output layer, is the actual output of the ith node in the output layer, is the regularization coefficient, is the weight from the hidden layer to the output layer, is the correlation coefficient, is the correlation coefficient between output node i and output node k, is the expected output of the kth node in the output layer, is the actual output of the kth node in the output layer;

[0095] The elite retention strategy is introduced, that is, after each iteration, a certain proportion of individuals with high fitness are selected from the population as elite individuals, without performing crossover genetic operations or mutation genetic operations, and directly retaining them intact to the next generation. Specifically:

[0096] ,

[0097] ,

[0098] In the formula, Reserve proportions for newer elites, To reserve a proportion for the basic elite, Reserve ratio for maximum elite, is the improvement in fitness, is the preset improvement threshold, To adjust the parameters;

[0099] According to the dynamic parameter adjustment strategy, the crossover rate and mutation rate are adjusted. The expression of the dynamic parameter adjustment strategy is:

[0100] ,

[0101] ,

[0102] In the formula, is the crossover probability, is the maximum crossover probability, is the minimum crossover probability, is the minimum number of iterations, is the fitness value with the maximum fitness among the two crossover individuals, is the average fitness value of the population, is the current iteration number, is the hyperbolic tangent function, is the minimum fitness value in the population, is the maximum fitness value in the population, is the fitness value of the mutant individual, is the maximum mutation probability, is the mutation probability, is the minimum mutation probability;

[0103] Set the minimum number of iterations , the convergence condition is introduced as the condition for the termination of iteration. If the convergence condition is met and the number of iterations is greater than the minimum number of iterations, the program ends and the optimal weight and threshold are output. The convergence condition is:

[0104] ,

[0105] In the formula, is the optimal fitness value of the tth generation, is the convergence threshold, is the optimal fitness value of the t-1th generation.

[0106] Step S105, inputting the acquired real-time rail vibration signal into the track defect monitoring model, and the track defect monitoring model outputs a monitoring result corresponding to the real-time rail vibration signal.

[0107] See also Figure 2 , which shows a structural block diagram of a track defect monitoring system based on an improved BP neural network of the present application.

[0108] like Figure 2 As shown, the track defect monitoring system 200 includes an acquisition module 210 , a decomposition module 220 , a fusion module 230 , a training module 240 and an output module 250 .

[0109] The acquisition module 210 is configured to acquire the rail vibration signal according to a preset signal anti-aliasing processing rule, and the expression of the signal anti-aliasing processing rule is:

[0110] ,

[0111] ,

[0112] In the formula, is the stop-band attenuation, is the Nyquist zone where the signal is located, is the sampling frequency, is the upper frequency limit of the signal, is the carrier frequency, For the integral, is the bandwidth of the collected rail vibration signal, is the 3dB bandwidth, is the lower frequency limit of the signal, is the filter order;

[0113] The decomposition module 220 is configured to perform VMD decomposition on the rail vibration signal, and filter the sample entropy of the IMF signal obtained by the decomposition according to the entropy value filtering feature matrix strategy to obtain a sample entropy feature matrix, wherein the sample entropy feature matrix contains the sample entropy features of each target IMF signal;

[0114] A fusion module 230 is configured to extract feature information from the rail vibration signal according to a pre-built DRSN deep learning model, and fuse the feature information with sample entropy features of each target IMF signal to obtain fused feature information;

[0115] A training module 240 is configured to input the fusion feature information into an improved BP neural network, and iteratively train the improved BP neural network based on a dynamic parameter adjustment strategy to obtain a track disease monitoring model;

[0116] The output module 250 is configured to input the acquired real-time rail vibration signal into the track defect monitoring model, and the track defect monitoring model outputs a monitoring result corresponding to the real-time rail vibration signal.

[0117] It should be understood that Figure 2 Modules and references documented in Figure 1 Therefore, the operations and features described above for the method and the corresponding technical effects are also applicable to Figure 2 The modules in it will not be described in detail here.

[0118] In some other embodiments, the embodiments of the present invention further provide a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor is caused to execute the track defect monitoring method based on the improved BP neural network in any of the above method embodiments;

[0119] As an implementation mode, the computer-readable storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are configured as follows:

[0120] The rail vibration signal is obtained according to the preset signal anti-aliasing processing rule, and the expression of the signal anti-aliasing processing rule is:

[0121] ,

[0122] ,

[0123] In the formula, is the stop-band attenuation, is the Nyquist zone where the signal is located, is the sampling frequency, is the upper frequency limit of the signal, is the carrier frequency, For the integral, is the bandwidth of the collected rail vibration signal, is the 3dB bandwidth, is the lower frequency limit of the signal, is the filter order;

[0124] Perform VMD decomposition on the rail vibration signal, and screen the sample entropy of the IMF signal obtained by the decomposition according to the entropy value screening feature matrix strategy to obtain a sample entropy feature matrix, wherein the sample entropy feature matrix contains the sample entropy features of each target IMF signal;

[0125] Extracting feature information from the rail vibration signal according to a pre-built DRSN deep learning model, and fusing the feature information with sample entropy features of each target IMF signal to obtain fused feature information;

[0126] Inputting the fused feature information into an improved BP neural network, and iteratively training the improved BP neural network based on a dynamic parameter adjustment strategy to obtain a track disease monitoring model;

[0127] The acquired real-time rail vibration signal is input into the track defect monitoring model, and the track defect monitoring model outputs a monitoring result corresponding to the real-time rail vibration signal.

[0128] The computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the track defect monitoring system based on the improved BP neural network, etc. In addition, the computer-readable storage medium may include a high-speed random access memory, and may also include a memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the track defect monitoring system based on the improved BP neural network via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0129] Figure 3 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 3 As shown, the device includes: a processor 310 and a memory 320. The electronic device may also include: an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330 and the output device 340 may be connected via a bus or other means. Figure 3 In the example, the connection through the bus is taken as an example. The memory 320 is the above-mentioned computer-readable storage medium. The processor 310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 320, that is, the track defect monitoring method based on the improved BP neural network in the above-mentioned method embodiment is realized. The input device 330 can receive input digital or character information, and generate key signal input related to user settings and function control of the track defect monitoring system based on the improved BP neural network. The output device 340 may include display devices such as display screens.

[0130] The electronic device can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0131] As an implementation mode, the electronic device is applied to a rail defect monitoring system based on an improved BP neural network, and is used for a client, and includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can:

[0132] The rail vibration signal is obtained according to the preset signal anti-aliasing processing rule, and the expression of the signal anti-aliasing processing rule is:

[0133] ,

[0134] ,

[0135] In the formula, is the stop-band attenuation, is the Nyquist zone where the signal is located, is the sampling frequency, is the upper frequency limit of the signal, is the carrier frequency, For the integral, is the bandwidth of the collected rail vibration signal, is the 3dB bandwidth, is the lower frequency limit of the signal, is the filter order;

[0136] Perform VMD decomposition on the rail vibration signal, and screen the sample entropy of the IMF signal obtained by the decomposition according to the entropy value screening feature matrix strategy to obtain a sample entropy feature matrix, wherein the sample entropy feature matrix contains the sample entropy features of each target IMF signal;

[0137] Extracting feature information from the rail vibration signal according to a pre-built DRSN deep learning model, and fusing the feature information with sample entropy features of each target IMF signal to obtain fused feature information;

[0138] Inputting the fused feature information into an improved BP neural network, and iteratively training the improved BP neural network based on a dynamic parameter adjustment strategy to obtain a track disease monitoring model;

[0139] The acquired real-time rail vibration signal is input into the track defect monitoring model, and the track defect monitoring model outputs a monitoring result corresponding to the real-time rail vibration signal.

[0140] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A track disease monitoring method based on improved BP neural network, characterized in that: include: The rail vibration signal is obtained according to the preset signal anti-aliasing processing rule, and the expression of the signal anti-aliasing processing rule is: ， ， In the formula, is the stop-band attenuation, is the Nyquist zone where the signal is located, is the sampling frequency, is the upper frequency limit of the signal, is the carrier frequency, For the integral, is the bandwidth of the collected rail vibration signal, is the 3dB bandwidth, is the lower frequency limit of the signal, is the filter order; Perform VMD decomposition on the rail vibration signal, and screen the sample entropy of the IMF signal obtained by the decomposition according to the entropy value screening feature matrix strategy to obtain a sample entropy feature matrix, wherein the sample entropy feature matrix contains the sample entropy features of each target IMF signal; Extracting feature information from the rail vibration signal according to a pre-built DRSN deep learning model, and fusing the feature information with sample entropy features of each target IMF signal to obtain fused feature information; Inputting the fused feature information into an improved BP neural network, and iteratively training the improved BP neural network based on a dynamic parameter adjustment strategy to obtain a track disease monitoring model; The acquired real-time rail vibration signal is input into the track defect monitoring model, and the track defect monitoring model outputs a monitoring result corresponding to the real-time rail vibration signal.

2. A track defect monitoring method based on improved BP neural network according to claim 1, characterized in that: The performing VMD decomposition on the rail vibration signal comprises: The rail vibration signal is modally decomposed to obtain each IMF signal, and each IMF signal is Hilbert transformed to obtain an analytical signal of each IMF signal. The expression of the single-sided spectrum of the analytical signal is: ， In the formula, is the impulse function, is an imaginary unit, is the kth IMF signal after decomposition, For time; By mixing an exponential term , the center frequency of each IMF signal is estimated, and the spectrum of each IMF signal obtained after decomposition is converted to the corresponding baseband, wherein the conversion expression is: ， In the formula, is the center frequency corresponding to each IMF signal after decomposition; The bandwidth of each IMF signal is estimated using the Gaussian smoothed offset signal, and the variational problem is constructed as follows: ， In the formula, is the original signal, It means to find the partial derivative of t. is the kth IMF signal after decomposition, is the center frequency corresponding to each IMF signal after decomposition; To solve the variational problem, a quadratic penalty factor is introduced and the Lagrange multiplication operator , the constrained variational problem is transformed into an unconstrained variational problem. The expression of the unconstrained variational problem is: ， In the formula, is the Lagrangian operation equation, is the Lagrange multiplication operator, is the input signal; After multiple iterative calculations, the minimum value of the Lagrangian expression is searched to obtain the optimal solution and the center frequency of the optimal solution. The iterative update expression is: ， ， ， In the formula, is the frequency expression of the kth signal at the n+1th iteration, is the frequency expression of the original signal, is the frequency expression of the ith signal, is the frequency expression of the Lagrangian operator, is the center frequency of the kth signal of the n+1th iteration, is the frequency variable, is the frequency expression of the kth signal, is the time constant, is the Lagrangian operator of the n+1th iteration, is the Lagrangian operator of the nth iteration; Given a discriminant accuracy , if the preset iteration condition is met, the iteration is stopped, where the expression of the iteration condition is: ， In the formula, is the kth signal of the n+1th iteration, is the kth signal of the nth iteration.

3. The track defect monitoring method based on improved BP neural network according to claim 1 is characterized in that: The strategy of screening the characteristic matrix according to the entropy value screens the sample entropy of the IMF signal obtained by decomposition, and the sample entropy characteristic matrix obtained includes: The IMF signal obtained by VMD decomposition is embedded into a high-dimensional space, and the vector Refactoring is performed, and the refactored expression is: ， In the formula, is the reconstructed signal, for The first vector of the vector, for The second vector of the vector, for The mth vector of a vector, for The first element of the kth vector of a vector, for The second element of the kth vector of a vector, for The mth element of the kth vector of a vector; For each vector in the embedding space, find the vector All other vectors under the same condition, so that the distance between all other vectors is less than a pre-set threshold ; Calculate the sample entropy of the IMF signal after VMD decomposition. The calculation formula is: ， In the formula, is the sample entropy, is the length of the time series, are the remaining vectors and vectors in the N-dimensional phase space The distance is greater than the threshold The probability of are the remaining vectors and vectors in the N-dimensional phase space The distance is less than the threshold probability; Taking the sample with the least IMF signal as a benchmark, the IMF signals with entropy values ​​greater than the first entropy threshold and less than the second entropy threshold are eliminated to form a sample entropy feature matrix.

4. The track defect monitoring method based on improved BP neural network according to claim 1 is characterized in that: The iterative training of the improved BP neural network based on the dynamic parameter adjustment strategy includes: Genetic algorithm is introduced to optimize the weights and thresholds of BP neural network. The initial weights and thresholds generated by BP neural network are used as the initial population of genetic algorithm. Real number individual coding is adopted. The coding method is as follows: ， Where L is the encoding length, n is the number of neurons in the input layer, and h is the number of neurons in the hidden layer. is the number of neurons in the output layer; The fitness function of the genetic algorithm is determined as: ， In the formula, is the fitness function, is the weight coefficient of the i-th output node, is the expected output of the i-th node in the output layer, is the actual output of the i-th node in the output layer, is the regularization coefficient, is the weight from the hidden layer to the output layer, is the correlation coefficient, is the correlation coefficient between output node i and output node k, is the expected output of the kth node in the output layer, is the actual output of the kth node in the output layer; The elite retention strategy is introduced, that is, after each iteration, a certain proportion of individuals with high fitness are selected from the population as elite individuals, without performing crossover genetic operations or mutation genetic operations, and directly retaining them intact to the next generation. Specifically: ， ， In the formula, Reserve proportions for newer elites, To reserve a proportion for the basic elite, Reserve ratio for maximum elite, is the improvement in fitness, is the preset improvement threshold, To adjust the parameters; According to the dynamic parameter adjustment strategy, the crossover rate and mutation rate are adjusted; Set the minimum number of iterations , introduce the convergence condition as the condition for the termination of iteration. If the convergence condition is met and the number of iterations is greater than the minimum number of iterations, the program ends and the optimal weight and threshold are output. The convergence condition is: ， In the formula, is the optimal fitness value of the tth generation, is the convergence threshold, is the optimal fitness value of the t-1th generation.

5. A track defect monitoring method based on improved BP neural network according to claim 4, characterized in that: The expression of the dynamic parameter adjustment strategy is: ， ， In the formula, is the crossover probability, is the maximum crossover probability, is the minimum crossover probability, is the minimum number of iterations, is the fitness value with the maximum fitness among the two crossover individuals, is the average fitness value of the population, is the current iteration number, is the hyperbolic tangent function, is the minimum fitness value in the population, is the maximum fitness value in the population, is the fitness value of the mutant individual, is the maximum mutation probability, is the mutation probability, is the minimum mutation probability.

6. The track defect monitoring method based on improved BP neural network according to claim 1 is characterized in that: The loss function of the improved BP neural network is a sample entropy loss function based on VMD decomposition, and the expression of the sample entropy loss function is: ， In the formula, is the sample entropy loss function, is the number of samples in the dataset, is the probability that sample a belongs to category B, is a binary number, which is 1 if sample a belongs to category B, otherwise it is 0.

7. A track defect monitoring system based on improved BP neural network, characterized in that: include: The acquisition module is configured to acquire the rail vibration signal according to a preset signal anti-aliasing processing rule, wherein the expression of the signal anti-aliasing processing rule is: ， ， In the formula, is the stop-band attenuation, is the Nyquist zone where the signal is located, is the sampling frequency, is the upper frequency limit of the signal, is the carrier frequency, For the integral, is the bandwidth of the collected rail vibration signal, is the 3dB bandwidth, is the lower frequency limit of the signal, is the filter order; A decomposition module is configured to perform VMD decomposition on the rail vibration signal, and screen the sample entropy of the IMF signal obtained by the decomposition according to the entropy value screening feature matrix strategy to obtain a sample entropy feature matrix, wherein the sample entropy feature matrix contains the sample entropy features of each target IMF signal; A fusion module is configured to extract feature information from the rail vibration signal according to a pre-built DRSN deep learning model, and fuse the feature information with the sample entropy features of each target IMF signal to obtain fused feature information; A training module is configured to input the fusion feature information into an improved BP neural network, and iteratively train the improved BP neural network based on a dynamic parameter adjustment strategy to obtain a track disease monitoring model; The output module is configured to input the acquired real-time rail vibration signal into the track defect monitoring model, and the track defect monitoring model outputs a monitoring result corresponding to the real-time rail vibration signal.

8. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images