Rail vibration acceleration inversion method, device and storage medium

Abstract

The application discloses a rail vibration acceleration inversion method and device and a storage medium. The inversion method comprises the following steps: extracting features of vibration acceleration signals of a track slab and a rail to obtain a first vibration feature sequence of the track slab and a vibration feature sequence of the rail; calculating a first correlation coefficient between the vibration feature sequence of the rail and the first vibration feature sequence after each movement, and determining a second vibration feature sequence according to the first correlation coefficient; performing modal decomposition on the second vibration feature sequence to obtain a plurality of modal components; calculating a second correlation coefficient between each modal component and the vibration feature sequence of the rail, and determining a third vibration feature sequence according to the second correlation coefficient; constructing and training an inversion model; and inverting the rail vibration acceleration by using the target inversion model and the measured track slab vibration acceleration. The application realizes long-term, real-time and safe non-contact monitoring of the rail vibration acceleration.

Description

Technical Field

[0001] This invention belongs to the field of rail vibration signal monitoring technology, and particularly relates to a rail vibration acceleration inversion method, equipment and storage medium. Background Technology

[0002] The essential characteristic of railway transportation is the interaction between wheels and rails. With the increasing speed of high-speed rail, the friction, compression, and impact effects on the rails become increasingly significant, and the factors affecting track safety become more complex, leading to a greater possibility of contact fatigue on the track surface. In various industrial fields, vibration signal measurement and analysis have become important means of fault detection and prediction. How to develop a novel, real-time, long-term, and safe rail condition monitoring system based on vibration signal characteristics has become an important research topic in the field of high-speed railway operation safety monitoring.

[0003] Traditional real-time monitoring methods involve placing sensors on rails or vehicles. Directly mounting sensors on rails can cause false detections by flaw detection vehicles and affect the efficiency of flaw detection. Furthermore, sensor detachment can affect train safety. Moreover, existing high-speed railways generally do not allow sensors to be installed on rails for extended periods, as the resulting safety hazards are unpredictable.

[0004] Existing track monitoring methods are relatively abundant and mature. However, to ensure train operation safety, high-speed railways generally do not allow sensors to be installed on the rails for extended periods. Therefore, rail vibration monitoring is currently limited to short- to medium-term monitoring and subsequent data analysis. Long-term, real-time, and safe monitoring of rail vibration acceleration has become a key challenge in railway infrastructure operation and maintenance. When trains run on the track, the impact between the wheels and rails causes rail vibration. These vibration waves are transmitted through the rails to the surrounding track slabs, inevitably resulting in a transmission effect of vertical vibration acceleration between the rails and track slabs. Furthermore, installing sensors on the track slabs offers advantages such as convenience and safety. Therefore, installing vibration sensors on the track slabs to achieve long-term, real-time, and safe non-contact monitoring of rail vertical vibration acceleration is feasible. Summary of the Invention

[0005] The purpose of this invention is to provide a method, device and storage medium for inverting rail vibration acceleration, so as to solve the problem that it is impossible to monitor the condition of rail in real time and for a long time because vibration sensors cannot be installed on rails for a long time.

[0006] This invention solves the above-mentioned technical problems through the following technical solution: a method for inverting rail vibration acceleration, the inversion method comprising the following steps:

[0007] Acquire vibration acceleration signals of the track slab and the rail;

[0008] Feature extraction is performed on the vibration acceleration signals of the track slab and the rail to obtain the first vibration feature sequence of the track slab and the vibration feature sequence of the rail.

[0009] The first vibration feature sequence is moved multiple times, and after each move, the first correlation coefficient between the vibration feature sequence of the rail and the first vibration feature sequence after the move is calculated. The first vibration feature sequence after the move corresponding to the maximum value of the first correlation coefficient is used as the second vibration feature sequence of the track slab.

[0010] Modal decomposition is performed on the second vibration feature sequence to obtain multiple modal components;

[0011] Calculate the second correlation coefficient between each modal component and the vibration characteristic sequence of the rail, and take the modal component corresponding to the maximum value of the second correlation coefficient as the third vibration characteristic sequence of the track slab;

[0012] An inversion model is constructed, with the third vibration feature sequence as the input and the rail vibration feature sequence as the output. The inversion model is then trained to obtain the target inversion model.

[0013] The track slab vibration acceleration is obtained in real time, and the rail vibration acceleration is inverted based on the track slab vibration acceleration and the target inversion model.

[0014] Furthermore, feature extraction is performed on the track slab vibration acceleration signal and the rail vibration acceleration signal. The specific implementation process includes:

[0015] The vibration acceleration signal of the track slab is denoised.

[0016] Endpoint detection is performed on the denoised track slab vibration acceleration signal, and each vibration feature segment of the track slab is extracted based on the detected endpoints; the vibration feature segments of the track slab are spliced ​​together to obtain the first vibration feature sequence of the track slab.

[0017] Based on the detected endpoints, extract each vibration characteristic segment of the rail, and splice the vibration characteristic segments of the rail to obtain the vibration characteristic sequence of the rail.

[0018] Furthermore, the vibration acceleration signal of the track slab is subjected to noise reduction processing, the specific implementation process of which is as follows:

[0019] The cutoff frequency of the Butterworth low-pass filter is set to 1Hz, and the trend signal is extracted from the track slab vibration acceleration signal using the Butterworth low-pass filter.

[0020] The track slab vibration acceleration signal is subtracted from the corresponding trend signal to obtain the denoised track slab vibration acceleration signal.

[0021] Furthermore, a dual-threshold method is used to perform endpoint detection on the denoised track slab vibration acceleration signal. The specific implementation process is as follows:

[0022] A sliding window is used to frame the denoised track slab vibration acceleration signal, resulting in multiple frames with overlap between adjacent frames.

[0023] The short-time zero-crossing rate and short-time energy of each frame of the signal are calculated using the following formulas:

[0024]

[0025] ,

[0026] in, Let N be the short-time zero-crossing rate of the i-th frame signal, and N be the number of vibration accelerations contained in the i-th frame signal. Let n be the value of the vibration acceleration in the i-th frame of the signal. It is a symbolic function; Let be the short-time energy of the signal in the i-th frame;

[0027] Set the first threshold value Based on the envelope of the short-time energy and the first threshold value Two intersection points have been preliminarily determined. , ;

[0028] Set a second threshold value Based on the envelope of the short-time energy and the second threshold value Determine the two intersection points , Among them, the intersection point Located at the intersection On the left, intersection Located at the intersection On the right side, ;

[0029] Set a third threshold value Based on the short-time zero-crossing rate, from the intersection point To the left, from the intersection Searching to the right, find those where the short-term zero-crossing rate is lower than the third threshold. two points , ,point , These are the two endpoints of the vibration characteristic segment of the track slab.

[0030] Furthermore, the first threshold value is specifically as follows: ,in The short-time energy average value of the silent signal is given. The silent signal includes the no-load noise signal and the vertical vibration acceleration signal of this track when a car on the adjacent track passes by.

[0031] The second threshold value is as follows: ;

[0032] The third threshold value is specifically: ,in, This represents the maximum value of the cross-correlation coefficient between frame signals.

[0033] Furthermore, the inversion model is a Transformer model, which includes a mask padding layer, a position encoder, an encoding layer, and a fully connected layer connected in sequence; the encoding layer includes a Multi-Head Attention layer, an Add & Norm layer, a Feed Forward layer, and an Add & Norm layer connected in sequence.

[0034] Based on the same concept, the present invention also provides an electronic device, the device comprising:

[0035] Memory, used to store computer programs;

[0036] A processor is used to execute the computer program to implement the steps of the rail vibration acceleration inversion method as described above.

[0037] Based on the same concept, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the rail vibration acceleration inversion method as described above.

[0038] Beneficial effects

[0039] Compared with the prior art, the advantages of the present invention are as follows:

[0040] This invention first extracts the corresponding vibration feature sequences from the rail vibration acceleration signal and the track slab vibration acceleration signal, respectively. Then, it moves the track slab vibration feature sequence to determine the track slab vibration feature sequence with the highest correlation coefficient to the rail vibration feature sequence. The VMD algorithm is then used to perform mode decomposition on the track slab vibration feature sequence to search for the mode component with the highest correlation coefficient to the rail vibration feature sequence. Next, an inversion model is used to mine the spatiotemporal data feature transfer characteristics between the rail vibration feature sequence and the track slab vibration feature sequence with the highest correlation coefficient, resulting in a target inversion model. Finally, the target inversion model and the measured track slab vibration acceleration signal are used to invert the rail vibration acceleration. This invention has good inversion results.

[0041] Vibration sensors are permanently deployed on the track slab. By using the track slab vibration acceleration signals measured by the vibration sensors and the target inversion model, the rail vibration acceleration can be inverted or predicted. This enables long-term, real-time, and safe non-contact monitoring of rail vibration acceleration, ensuring train operation safety. Attached Figure Description

[0042] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of the rail vibration acceleration inversion method in this embodiment of the invention;

[0044] Figure 2 This is a schematic diagram of the vertical vibration acceleration signal of the original track slab in an embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram of the trend signal of the vertical vibration acceleration signal of the track slab in an embodiment of the present invention;

[0046] Figure 4 This is a schematic diagram of the vertical vibration acceleration signal of the track slab after removing baseline drift in an embodiment of the present invention;

[0047] Figure 5(a) shows the measured vertical vibration acceleration signal (including the silent signal and the vertical vibration acceleration signal) in an embodiment of the present invention.

[0048] Figure 5(b) is a schematic diagram of the vibration signal of the adjacent pillow in an embodiment of the present invention;

[0049] Figure 6 This is a schematic diagram of the vibration characteristic segment of the track slab in an embodiment of the present invention;

[0050] Figure 7(a) is a schematic diagram of the vibration characteristic sequence obtained by splicing together the vibration characteristic segments of the rail in an embodiment of the present invention;

[0051] Figure 7(b) is a schematic diagram of the vibration feature sequence obtained by splicing together the vibration feature segments of the track slab in an embodiment of the present invention;

[0052] Figure 8 This is a flowchart of the time lag cross-correlation calculation in an embodiment of the present invention;

[0053] Figure 9 This is the result of calculating the first correlation coefficient (Pearson correlation coefficient) in this embodiment of the invention;

[0054] Figure 10(a) is a VMD decomposition waveform diagram in an embodiment of the present invention;

[0055] Figure 10(b) is a spectrum diagram of each modal component in an embodiment of the present invention;

[0056] Figure 11 This is the second correlation coefficient heatmap after VMD decomposition in this embodiment of the invention;

[0057] Figure 12(a) is a waveform diagram of the measured track slab vibration acceleration in an embodiment of the present invention;

[0058] Figure 12(b) is a waveform diagram of the measured rail vibration acceleration in an embodiment of the present invention;

[0059] Figure 12(c) is a waveform diagram of the inverted rail vibration acceleration in an embodiment of the present invention. Detailed Implementation

[0060] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0062] like Figure 1 As shown in this embodiment, a rail vibration acceleration inversion method includes the following steps:

[0063] Step 1: Obtain the vibration acceleration signal of the track slab and the vibration acceleration signal of the rail.

[0064] The rail vibration acceleration signal and track slab vibration acceleration signal in this embodiment are from a certain intercity high-speed railway in western China. The line is about 70km long and has a design speed of 250km / h. The entire line is laid with CRTS III type slab track without ballast. The signal acquisition site is the roadbed section of the line.

[0065] This embodiment takes vertical vibration acceleration as an example. To obtain the vertical vibration acceleration signals of the rail and the track slab, vibration sensors are evenly distributed between the rail and the track slab between two sleepers. These sensors collect the corresponding vertical vibration acceleration signals. In this embodiment, the vibration sensor on the rail has a range of 100g and is used to collect the rail's vertical vibration acceleration signal, while the vibration sensor on the track slab has a range of 10g and is used to collect the track slab's vertical vibration acceleration signal. To protect the vibration sensors, a protective casing is installed around them. The vibration sensors are connected to a signal processing device via a data cable, which then performs the inversion of the rail's vertical vibration acceleration. The data cable is housed inside a rubber tube for protection.

[0066] Step 2: Extract features from the track slab vibration acceleration signal and the rail vibration acceleration signal to obtain the first vibration feature sequence of the track slab and the vibration feature sequence of the rail.

[0067] In this embodiment, feature extraction is performed on the track slab vibration acceleration signal and the rail vibration acceleration signal. The specific implementation process is as follows:

[0068] Step 2.1: Denoise the track slab vibration acceleration signal;

[0069] Step 2.2: Perform endpoint detection on the denoised track slab vibration acceleration signal, and extract each vibration feature segment of the track slab based on the detected endpoints; splice the vibration feature segments of the track slab to obtain the first vibration feature sequence of the track slab.

[0070] Step 2.3: Extract each vibration feature segment of the rail based on the detected endpoints, and splice the vibration feature segments of the rail to obtain the vibration feature sequence of the rail.

[0071] Due to the influence of low-frequency noise from the vibration sensor, environmental noise at the data acquisition site, and initial vibration acceleration, the vertical vibration acceleration signal acquired by the vibration sensor exhibits a severe baseline shift, such as... Figure 2 The vertical vibration acceleration signal of the track slab shown is... Figure 2 The ordinate represents the vertical vibration acceleration. Therefore, the acquired vertical vibration acceleration signal needs to be denoised. In step 2.1, this embodiment uses a Butterworth low-pass filter for denoising, and the specific implementation process is as follows:

[0072] Step 2.11: Set the cutoff frequency of the Butterworth low-pass filter to 1Hz, and use the Butterworth low-pass filter to extract the trend signal from the track slab vibration acceleration signal.

[0073] Step 2.12: Subtract the track slab vibration acceleration signal from the corresponding trend signal to obtain the denoised track slab vibration acceleration signal.

[0074] Baseline drift signals are typically located at low frequencies. Since the vibration frequency of the trend signal is less than 1 Hz, setting the cutoff frequency of the Butterworth low-pass filter to 1 Hz allows the trend signal to be extracted from the vertical vibration acceleration signal of the track slab. Figure 3 The trend signal of the track slab extracted using a Butterworth low-pass filter is shown. Figure 3 The vertical axis represents the vertical vibration acceleration.

[0075] Subtracting the trend signal from the original signal yields the detrended signal (i.e., the denoised vertical vibration acceleration signal). For example, subtracting the trend signal of the track slab from the vertical vibration acceleration signal of the track slab results in the denoised vertical vibration acceleration signal of the track slab. Figure 4 The vertical vibration acceleration signal of the track slab after destressing is shown. Figure 4 The vertical axis represents the vertical vibration acceleration.

[0076] The vertical vibration acceleration signal measured by the vibration sensor on the track includes not only the vertical vibration acceleration signal but also the silent signal. The silent signal includes the track no-load noise signal and the vertical vibration acceleration signal of the track when a train passes over the adjacent track (i.e., the vertical vibration acceleration signal of the steel plate of the track and the vertical vibration acceleration signal of the track slab caused by the adjacent track, referred to as the adjacent sleeper vibration signal), as shown in Figures 5(a) and 5(b). The silent signal accounts for more than 90% of the total measured signal. The vibration sensor is always in the acquisition state. The vibration sensor also acquires signals when no train passes over the track. The signal acquired at this time is the no-load noise signal.

[0077] To filter out silent signals and extract useful vibration feature sequences for subsequent processing, this embodiment employs a dual-threshold method to detect the endpoints of the denoised track slab's vertical vibration acceleration signal. The purpose of this invention is to estimate the rail's vertical vibration acceleration signal by inverting the track slab's vertical vibration acceleration signal. Therefore, only each vibration feature segment of the track slab needs to be extracted. Since the vibration feature segments of both the rail and track slab are acquired simultaneously, the time range corresponding to the obtained track slab vibration feature segments is the same as that of the rail vibration feature segments. The dual-threshold method can effectively identify the track's vibration feature segments. The dual-threshold method can extract each vibration feature segment based on the different characteristics of the vertical vibration acceleration signal compared to silent and interference signals. These vibration feature segments are then compared with a set threshold value to achieve the purpose of vibration endpoint detection.

[0078] In step 2.2, the dual-threshold method is used to perform endpoint detection on the denoised track slab vertical vibration acceleration signal. The specific implementation process is as follows:

[0079] Step 2.21: Use a sliding window to frame the denoised vertical vibration acceleration signal of the track slab, resulting in multiple frames with overlap between adjacent frames.

[0080] The size of the sliding window remains constant, so the frame length of each frame of signal is the same. In this embodiment, the overlap is half a frame length.

[0081] Step 2.22: Calculate the short-time zero-crossing rate and short-time energy of each frame of signal. The specific calculation formula is as follows:

[0082] (1)

[0083] (2)

[0084] (3)

[0085] in, Let N be the short-time zero-crossing rate of the i-th frame signal, and N be the number of vertical vibration accelerations contained in the i-th frame signal (i.e., the frame length). Let n be the value of the nth vertical vibration acceleration in the i-th frame of the signal. It is a symbolic function; Let be the short-time energy of the i-th frame signal.

[0086] Step 2.23: Set the first threshold value Based on the envelope of the short-time energy and the first threshold value Two intersection points have been preliminarily determined. , .

[0087] Select a first threshold value on the short-time energy envelope. Higher than the first threshold value This must be vibration acceleration, and the start and end points of the vibration characteristic segment should be located at the first threshold value. Intersection with the short-time energy envelope , Outside of the corresponding time period. In this embodiment, ,in This represents the short-time energy average of the silent signal.

[0088] Step 2.24: Set the second threshold value Based on the envelope of the short-time energy and the second threshold value Determine the two intersection points , Among them, the intersection point Located at the intersection On the left, intersection Located at the intersection On the right side, .

[0089] Select a second threshold value on the short-time energy envelope. , and from the intersection To the left, intersection Searching to the right, we find the short-term energy envelope and the second threshold value respectively. Two intersecting points , , This represents the vibration characteristic segment initially determined by the dual-threshold method. In this embodiment, .

[0090] Step 2.25: Set the third threshold value Based on the short-time zero-crossing rate, from the intersection point To the left, from the intersection Searching to the right, find those where the short-term zero-crossing rate is lower than the third threshold. two points , ,point , These are the two endpoints of the vibration characteristic segment of the track slab, points... Point is the starting point of vibration. This is the end point of the vibration.

[0091] In this embodiment, ,in, This represents the maximum value of the cross-correlation coefficient between frame signals.

[0092] Similarly, the endpoints of each vibration characteristic segment of the track slab can be extracted, and then all vibration characteristic segments of the track slab can be determined based on the endpoints. Since the vibration characteristic segments of the rail and the track slab are collected simultaneously, the time range corresponding to the vibration characteristic segments of the track slab is the same as that corresponding to the vibration characteristic segments of the rail. Therefore, the vibration characteristic segments of the rail can be extracted based on the detected endpoints of the track slab, thereby allowing the extraction of all vibration characteristic segments of the rail.

[0093] In another specific embodiment of the present invention, the vibration characteristic sequence of the rail is extracted according to steps 2.1 to 2.3.

[0094] Figure 6 The extracted vibration characteristic segments of the track slab are shown. Figures 7(a) and 7(b) show the vibration characteristic sequences obtained by splicing the extracted vibration characteristic segments of the rail and track slab, respectively. Since the vibration of the track slab and the vibration of the rail occur almost simultaneously, the endpoint data of each vibration characteristic segment of the track slab extracted by the double-threshold method are substituted into the vertical vibration acceleration signal of the rail to obtain each vibration characteristic segment of the rail. The vibration characteristic segments of the rail are then spliced ​​to obtain the vibration characteristic sequence of the rail. Step 3: The first vibration characteristic sequence is moved multiple times, and after each move, the first correlation coefficient between the vibration characteristic sequence of the rail and the moved first vibration characteristic sequence is calculated.

[0095] In practical applications, since the vibration sensors placed manually may not be on the same cross-section (i.e., the vibration sensors on the rail and the track slab are not entirely on the same cross-section), and due to the transmission effect of vertical vibration, the vertical vibration acceleration of the track slab may have a certain lag effect relative to the rail, it is necessary to calculate the time lag cross correlation (TLCC). TLCC is measured by gradually shifting a time series vector and repeatedly calculating the correlation coefficient between the two signals. If the peak of the correlation coefficient is located in the center, it means that the two time series are most correlated at this time. However, if there is a certain deviation in the position of the vibration sensor, then it is necessary to shift one sensor signal, with the number of frames shifted ranging from -150 to 150.

[0096] Using the first vibration characteristic sequence obtained in step 2 as the initial sequence, the initial sequence is shifted to the left N1 times, and after each shift, the first correlation coefficient (Pearson correlation coefficient) between the shifted first vibration characteristic sequence and the rail vibration characteristic sequence is calculated, resulting in N1 Pearson correlation coefficients. The initial sequence is then shifted to the right N2 times, and after each shift, the first correlation coefficient (Pearson correlation coefficient) between the shifted first vibration characteristic sequence and the rail vibration characteristic sequence is calculated, resulting in N2 Pearson correlation coefficients. Then, the maximum value is found among the N1 + N2 Pearson correlation coefficients; the shifted first vibration characteristic sequence corresponding to this maximum value is the second vibration characteristic sequence of the track slab. Figure 8 As shown. Figure 9 The graph showing the relationship between frame shift magnitude and Pearson correlation coefficient is presented. Figure 9 It can be seen that when the initial sequence is shifted one frame to the right, the Pearson correlation coefficient between the first vibration characteristic sequence after the shift and the vibration characteristic sequence of the rail is the largest.

[0097] Step 4: Perform modal decomposition on the second vibration feature sequence to obtain multiple modal components.

[0098] Because the track slab has a more complex structure than the rail and its vibration acceleration amplitude is weaker, the inverted characteristic signal is easily obscured by interference noise. Generally, the inverted characteristic signal is distributed within a specific frequency range, but this range is difficult to define. Therefore, the VMD algorithm (Variational Mode Decomposition) is used to perform mode decomposition on the second vibration characteristic sequence to obtain multiple modal components. That is, the VMD algorithm decomposes the second vibration characteristic sequence of the track slab into signals in different frequency ranges, and this frequency range can be adjusted according to the number of input mode decompositions. Then, the second correlation coefficient between each modal component and the vibration characteristic sequence of the rail is calculated until the modal component with the largest second correlation coefficient with the vibration characteristic sequence of the rail is found.

[0099] In this embodiment, the VMD algorithm uses 10 modal decompositions with a penalty factor of 1000, and the decomposition results are shown in Figure 10. No obvious modal overlap is observed in Figure 10, achieving a relatively ideal effect.

[0100] Step 5: Calculate the second correlation coefficient between each modal component and the vibration characteristic sequence of the rail, and take the modal component corresponding to the maximum value of the second correlation coefficient as the third vibration characteristic sequence of the track slab.

[0101] Figure 11The heatmap shows the Pearson correlation coefficient (i.e., the second correlation coefficient) between each modal component and the rail vibration characteristic sequence, with the target being the rail vibration characteristic sequence. It can be found that the Pearson correlation coefficient between the IMF7 modal component and the rail vibration characteristic sequence is the largest. Subsequently, the IMF7 modal component is obtained by VMD decomposition and inversion estimation analysis is performed.

[0102] Step 6: Construct an inversion model. Use the third vibration feature sequence as the input of the inversion model and the vibration feature sequence of the rail as the output of the inversion model. Train the inversion model to obtain the target inversion model.

[0103] The inversion model can be an existing artificial intelligence model, such as a neural network or a support vector machine. This embodiment uses the Transformer model as the inversion model. The Transformer model includes a mask padding layer, a position encoder, an encoding layer, and a fully connected layer connected in sequence. The encoding layer includes a Multi-Head Attention layer, an Add & Norm layer, a Feed Forward layer, and an Add & Norm layer connected in sequence.

[0104] Single-head attention mechanisms utilize three different spatial vector matrices to perform linear transformations on the input vibration feature sequence. Multi-Head Attention layers, on the other hand, perform multiple independent linear transformation calculations, calculating different results for each dimension, and finally concatenating these results for the final output. This allows the model to focus on sequence information at different levels, increasing its learning ability. Add is a residual connection, frequently used when the number of layers is large (especially in ResNet). Norm refers to Layer Normalization, which improves convergence speed by ensuring that the inputs of neurons in each layer have the same mean and variance. The layer that translates the vibration feature sequence input into the model into the target time series data is called the decoding layer. Traditional Transformer decoders tend to accumulate errors in time series regression and inversion. Since the prediction in this invention is a regression and inversion process, a complex decoder structure is unnecessary. Therefore, this invention uses a fully connected layer to replace the original model's decoder layer.

[0105] In this embodiment, the third vibration characteristic sequence of the track slab and the vibration characteristic sequence of the rail each have 42,823 data points, and 1,000 data points are randomly selected as verification data to verify the effectiveness of the inversion model.

[0106] Transformer model parameter settings: 6 encoder layers, 512-dimensional weight matrix, 8-head multi-head attention layer, 128 nodes in the fully connected layer, MSE (mean squared error) loss function, 10 batch size, Adam optimizer, sigmoid activation function, and 1000 training epochs. The training results of the inversion model are shown in Figures 12(a) to 12(c). It can be clearly seen that the rail vibration acceleration estimated by the VMD-Transformer model of this invention and the measured rail vibration acceleration maintain a consistent vibration trend in the high-amplitude vibration range. The VMD-Transformer model can effectively estimate the measured rail vibration acceleration data, demonstrating strong peak feature extraction capabilities.

[0107] Step 7: Obtain the real-time monitored track slab vibration acceleration, and invert the rail vibration acceleration based on the real-time monitored track slab vibration acceleration and the target inversion model.

[0108] In practical applications, it is necessary to extract the vibration characteristic sequence from the real-time monitored track slab vibration acceleration, and then use the inversion model to estimate the rail vibration acceleration.

[0109] To evaluate the performance of the inversion model of this invention, the VMD-Transformer model of this invention is compared with the VMD-L model, the Transformer model, etc. The difference between the VMD-L model and the VMD-Transformer model is that the former uses LSTM as the deep learning model, while the latter uses the Transformer model as the deep learning model. Prediction accuracy is evaluated using root mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination R2 score.

[0110] Mean absolute error (MAE) is the average of the absolute values ​​of prediction errors, used to measure the absolute difference between predicted and true values. Root mean square error (RMSE) is the standard deviation of the squared prediction errors, used to measure the difference between predicted and true values. R² score, also known as the coefficient of determination, is an indicator used to measure the goodness of fit of the model, with a value ranging from 0 to 1. The prediction or inversion results of the VMD-Transformer model, VMD-L model, and Transformer model of this invention are shown in Table 1.

[0111] Table 1. Prediction or inversion results of different methods

[0112]

[0113] As shown in Table 1, the VMD-Transformer model of this invention has better inversion performance compared to the VMD-L model and the Transformer model.

[0114] This invention also provides an electronic device, which includes a processor and a memory storing a computer program, wherein the processor is configured to execute the computer program to implement the steps of the rail vibration acceleration inversion method as described above.

[0115] Although not shown, the electronic device includes a processor that can perform various appropriate operations and processes based on programs and / or data stored in read-only memory (ROM) or loaded from a storage portion into random access memory (RAM). The processor can be a multi-core processor or may include multiple processors. In some embodiments, the processor may include a general-purpose main processor and one or more specialized coprocessors, such as a central processing unit, graphics processing unit (GPU), neural network processor (NPU), digital signal processor (DSP), etc. Various programs and data required for the operation of the electronic device are also stored in the RAM. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0116] The processor and memory described above are used together to execute programs stored in the memory. When the program is executed by a computer, it can implement the methods, steps, or functions described in the above embodiments.

[0117] Although not shown, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the rail vibration acceleration inversion method as described above.

[0118] Storage media in embodiments of the present invention include articles that are permanent or non-permanent, removable or non-removable, and can store information by any method or technology. Examples of storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information that can be accessed by a computing device.

[0119] The above description only discloses specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or modifications that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for inverting rail vibration acceleration, characterized in that, The inversion method includes the following steps: Acquire vibration acceleration signals of the track slab and the rail; Feature extraction is performed on the vibration acceleration signals of the track slab and the rail to obtain the first vibration feature sequence of the track slab and the vibration feature sequence of the rail. Using the first vibration characteristic sequence as the initial sequence, the initial sequence is shifted to the left N1 times, and after each shift, the first correlation coefficient between the shifted first vibration characteristic sequence and the vibration characteristic sequence of the rail is calculated; the initial sequence is shifted to the right N2 times, and after each shift, the first correlation coefficient between the shifted first vibration characteristic sequence and the vibration characteristic sequence of the rail is calculated; the maximum value is found from the N1+N2 first correlation coefficients, and the shifted first vibration characteristic sequence corresponding to the maximum value is the second vibration characteristic sequence of the track slab; Modal decomposition is performed on the second vibration feature sequence to obtain multiple modal components; Calculate the second correlation coefficient between each modal component and the vibration characteristic sequence of the rail, and take the modal component corresponding to the maximum value of the second correlation coefficient as the third vibration characteristic sequence of the track slab; An inversion model is constructed, with the third vibration feature sequence as the input and the rail vibration feature sequence as the output. The inversion model is then trained to obtain the target inversion model. The track slab vibration acceleration is obtained in real time, and the rail vibration acceleration is inverted based on the track slab vibration acceleration and the target inversion model.

2. The rail vibration acceleration inversion method according to claim 1, characterized in that, Feature extraction is performed on the vibration acceleration signals of the track slab and the rail. The specific implementation process includes: The vibration acceleration signal of the track slab is denoised. Endpoint detection is performed on the denoised track slab vibration acceleration signal, and each vibration feature segment of the track slab is extracted based on the detected endpoints; the vibration feature segments of the track slab are spliced ​​together to obtain the first vibration feature sequence of the track slab. Based on the detected endpoints, extract each vibration characteristic segment of the rail, and splice the vibration characteristic segments of the rail to obtain the vibration characteristic sequence of the rail.

3. The rail vibration acceleration inversion method according to claim 2, characterized in that, The noise reduction process for the track slab vibration acceleration signal is as follows: The cutoff frequency of the Butterworth low-pass filter is set to 1Hz, and the trend signal is extracted from the track slab vibration acceleration signal using the Butterworth low-pass filter. The track slab vibration acceleration signal is subtracted from the corresponding trend signal to obtain the denoised track slab vibration acceleration signal.

4. The rail vibration acceleration inversion method according to claim 2, characterized in that, The dual-threshold method is used to perform endpoint detection on the denoised track slab vibration acceleration signal. The specific implementation process is as follows: A sliding window is used to frame the denoised track slab vibration acceleration signal, resulting in multiple frames with overlap between adjacent frames. The short-time zero-crossing rate and short-time energy of each frame of the signal are calculated using the following formulas: ， in, Let N be the short-time zero-crossing rate of the i-th frame signal, and N be the number of vibration accelerations contained in the i-th frame signal. Let n be the value of the vibration acceleration in the i-th frame of the signal. It is a symbolic function; Let be the short-time energy of the signal in the i-th frame; Set the first threshold value Based on the envelope of the short-time energy and the first threshold value Two intersection points have been preliminarily determined. , ; Set a second threshold value Based on the envelope of the short-time energy and the second threshold value Determine the two intersection points , Among them, the intersection point Located at the intersection On the left, intersection Located at the intersection On the right side, ; Set a third threshold value Based on the short-time zero-crossing rate, from the intersection point To the left, from the intersection Searching to the right, find those where the short-term zero-crossing rate is lower than the third threshold. two points , ,point , These are the two endpoints of the vibration characteristic segment of the track slab.

5. The rail vibration acceleration inversion method according to claim 4, characterized in that, The first threshold value is specifically: ,in The short-time energy average value of the silent signal is given. The silent signal includes the no-load noise signal and the vertical vibration acceleration signal of this track when a car on the adjacent track passes by. The second threshold value is as follows: ; The third threshold value is specifically: ,in, This represents the maximum value of the cross-correlation coefficient between frame signals.

6. The rail vibration acceleration inversion method according to any one of claims 1 to 5, characterized in that, The inversion model is a Transformer model, which includes a mask padding layer, a position encoder, an encoding layer, and a fully connected layer connected in sequence; the encoding layer includes a Multi-Head Attention layer, an Add & Norm layer, a Feed Forward layer, and an Add & Norm layer connected in sequence.

7. An electronic device, characterized in that, The device includes: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the rail vibration acceleration inversion method as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the rail vibration acceleration inversion method as described in any one of claims 1 to 6.

Images