High-speed train derailment monitoring method, device and system

Abstract

The invention provides a high-speed train derailment monitoring method, device and system, and relates to the technical field of train monitoring. Comprising the following steps: collecting seismic oscillation data of different earthquake magnitudes as vibration judgment data of a high-speed train impact beam, and dividing the vibration judgment data into different vibration intensities; constructing a sample data set by the preprocessed vibration judgment data, and dividing the sample data set into a training set and a test set; building a convolution-long and short time recursion series neural network model, and performing training and testing by using the training set and the testing set to obtain a trained vibration intensity prediction model; and acquiring a real-time vibration signal acquired by the optical fiber sensor, inputting the preprocessed real-time vibration signal into the vibration intensity prediction model, and outputting the predicted vibration intensity by the vibration intensity prediction model. After an accident occurs, the collision level can be quickly positioned, analyzed and predicted in the first time, an alarm is given, rescue is organized, meanwhile, the device can absorb impact kinetic energy, and casualties are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of train monitoring, and in particular to a method, device and system for monitoring derailment of a high-speed train. Background Art

[0002] With the continuous increase in train speed, train derailment accidents occur frequently due to human and natural factors. After a train derailment accident occurs, a series of emergency measures need to be taken quickly to ensure the safety of personnel, reduce losses, and conduct subsequent investigation and recovery work, including immediate rescue and first aid, evacuation and safe handling, on-site protection and recording, information notification, accident investigation and aftermath handling, etc., to minimize the losses caused by train derailment accidents and ensure that similar accidents are effectively prevented in the future.

[0003] The corresponding emergency measures for train derailment vary according to the severity of the accident. (1) In the case of a minor derailment, some carriages of the train slightly deviate from the track, there are no casualties, and the train remains basically stable. At this time, the emergency braking system needs to be used immediately to prevent the train from moving further, and the passengers need to be comforted, the track needs to be inspected, and the accident needs to be reported to the superior management department in a timely manner. (2) In the case of a moderate derailment, multiple carriages of the train derail, which may cause minor injuries and partial damage to the train. At this time, it is necessary to call for emergency rescue services immediately and provide on-site first aid; evacuate passengers in an orderly manner, set up a safe area to prevent unauthorized personnel from entering; record the accident scene, collect relevant evidence; and promptly report the accident to the public and superior management department. (3) In the case of a serious derailment, multiple carriages of the train derail seriously, which may cause major casualties and property losses, and the train is seriously damaged. At this time, large-scale rescue and first aid are needed, and all passengers and staff need to be evacuated quickly and orderly; the accident situation should be reported to the public through the media and social platforms to avoid the spread of rumors; an investigation team should be established to investigate the cause of the accident in detail; and gradual restoration and reconstruction should be carried out to deal with the aftermath. (4) When a major derailment occurs, multiple train carriages will be seriously derailed, causing a large number of casualties and significant property losses. The train will be completely damaged and may cause secondary disasters. At this time, a large amount of timely rescue and other handling work is required.

[0004] After the train derails, taking prompt measures is the only solution to prevent the accident from further expanding. The crash barriers in the prior art do not have intelligent alarm technology. When an accident occurs, they cannot quickly locate the accident and alarm to organize rescue. In addition, the prior art cannot determine the accident level in the first time of the collision, which will affect the organization of rescue efforts. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a high-speed train derailment monitoring method, device and system, which can predict the severity of the derailment collision accident as soon as it occurs, and send an alarm to the superior competent department to facilitate the organization of rescue. At the same time, the device can absorb the impact kinetic energy of the train, reduce the degree of damage to the train, and minimize casualties and property losses.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present application provides a high-speed train derailment monitoring method, comprising:

[0008] S1: Classifying vibration intensity levels: collecting earthquake motion data of different magnitudes as vibration determination data D1, and classifying the vibration determination data D1 into different vibration intensity levels according to the level of the vibration determination data D1.

[0009] S2: Establishing a sample data set: preprocessing the vibration determination data D1, constructing a sample data set D2 from the preprocessed vibration determination data D1, and dividing the sample data set D2 into a training set D3 and a test set D4.

[0010] S3: Establish a vibration intensity prediction model P: Build a convolution-long-short time recursive series neural network model, use the training set D3 and the test set D4 to train and test the convolution-long-short time recursive series neural network model, and obtain a trained vibration intensity prediction model P.

[0011] S4: Predicting the real-time impact intensity level: When an impact occurs, a real-time vibration signal D5 is collected by an optical fiber sensor, the real-time vibration signal D5 is preprocessed, and the preprocessed real-time vibration signal D5 is input into a vibration intensity prediction model P, which outputs a predicted vibration intensity D6.

[0012] S5: Send prediction results and alarm: send the predicted vibration intensity D6 to the disaster monitoring cloud platform so that the disaster monitoring cloud platform sends an alarm signal.

[0013] Preferably, in step S1, seismic waves of different magnitudes are selected and divided into impact waves of corresponding intensities; the seismic motion data of different magnitudes are seismic motion data in the magnitude range of Ms2.0-Ms7.4 selected from the database; and seismic waves with a recording time of at least 3 seconds and an epicenter within a preset longitude and latitude range are selected from the seismic motion data.

[0014] Preferably, the preprocessing work in step S2 includes: automatically picking up seismic waves using long-short time average and Akaike criterion to obtain the arrival time of the seismic waves in the vertical direction; screening and eliminating abnormal information of the picked up seismic waves; extracting sampling points from the eliminated seismic waves according to a preset sampling frequency to form an acceleration time history curve; and all the acceleration time history curves constitute a sample data set D2.

[0015] Preferably, the convolution-long-short time recursive series neural network model in step S3 includes an input layer, a convolution layer, a pooling layer, a recursive neural network layer, a fully connected layer and an output layer; the acceleration time-history curve and the corresponding vibration intensity in the training set D3 are input into the convolution-long-short time recursive series neural network model for training; the acceleration time-history curve in the test set D4 is input into the trained convolution-long-short time recursive series neural network model for filtering and sampling, and then the predicted vibration intensity is output; the mean square error is calculated according to the predicted vibration intensity and the actual vibration intensity in the test set D4. When the mean square error is less than a preset threshold, the training of the convolution-long-short time recursive series neural network model is completed, and the vibration intensity prediction model P is obtained.

[0016] In a second aspect, the present application also provides a high-speed train derailment monitoring device, including an acquisition module, a preprocessing module, a model building module, a prediction module and an alarm module.

[0017] Among them, the acquisition module collects seismic data of different magnitudes as vibration determination data D1, and divides the vibration data into different vibration intensities.

[0018] Preferably, the acquisition module includes a first selection unit: selecting seismic motion data in the magnitude range of Ms2.0-Ms7.4 from the database; a second selection unit: selecting seismic waves with a recording time of at least 3 seconds and an epicenter within a preset longitude and latitude range from the seismic motion data; and a division unit: dividing the selected seismic waves of different magnitudes into impact waves of corresponding vibration intensities.

[0019] Wherein, the preprocessing module preprocesses the vibration determination data D1, constructs a sample data set D2 from the preprocessed vibration determination data D1, and divides the sample data set D2 into a training set D3 and a test set D4.

[0020] Preferably, the preprocessing module includes: an automatic picking unit: using long-short time average and Akaike criterion to automatically pick up seismic waves to obtain the arrival time of the seismic waves in the vertical direction; a screening unit: screening the abnormal information of the picked seismic waves and eliminating them; a first sampling unit: extracting sampling points according to a preset sampling frequency in the seismic waves after elimination to form an acceleration time history curve; a data set construction unit: a sample data set is constructed by all the acceleration time history curves.

[0021] Among them, the model building module: builds a convolution-long-short time recursive series neural network model, uses the training set D3 and the test set D4 to train and test the convolution-long-short time recursive series neural network model, and obtains the trained vibration intensity prediction model P.

[0022] Preferably, the model building module includes: a model building unit: a convolution-long-short time recursive series neural network model is composed of an input layer, a convolution layer, a pooling layer, a recursive neural network layer, a fully connected layer and an output layer; a training unit: the acceleration time-history curve and the corresponding vibration intensity in the training set D3 are input into the convolution-long-short time recursive series neural network model for training; a testing unit: the acceleration time-history curve in the test set D4 is input into the trained convolution-long-short time recursive series neural network model for filtering and sampling, and then the predicted vibration intensity is output; a model correction unit: the mean square error is calculated according to the predicted vibration intensity and the actual vibration intensity in the test set D4. When the mean square error is less than a preset threshold, the convolution-long-short time recursive series neural network model training is completed to obtain a vibration intensity prediction model.

[0023] Among them, the prediction module: obtains the real-time vibration signal D5 collected by the optical fiber sensor set on the impact beam, preprocesses the real-time vibration signal D5, inputs the preprocessed real-time vibration signal D5 into the vibration intensity prediction model P, and the vibration intensity prediction model P outputs the predicted impact intensity D6.

[0024] Among them, the alarm module: sends the impact intensity D6 to the disaster monitoring cloud platform, so that the disaster monitoring cloud platform sends an alarm signal.

[0025] In a third aspect, the present application also provides a high-speed train derailment monitoring system, comprising:

[0026] The anti-impact system includes crash barriers arranged on both sides of the track, wherein the crash barriers include transversely arranged impact beams and vertically arranged crash barriers, the impact beams and crash barriers are fixedly connected via energy absorbing blocks, and a plurality of optical fiber sensors are arranged on the inner side of the impact beams.

[0027] The signal processing system is electrically connected to the optical fiber sensor, receives the real-time vibration signal D5 collected by the optical fiber sensor, and calculates the impact intensity D6 of the real-time vibration signal D5.

[0028] The local monitoring platform is used to receive and display the impact intensity D6 calculated by the signal processing system.

[0029] The disaster monitoring cloud platform is electrically connected to the signal processing system and the optical fiber sensor, receives the vibration position collected by the optical fiber sensor and the impact intensity calculated by the signal processing system, and sends an alarm signal.

[0030] Preferably, the optical fiber sensors are evenly arranged on the inner side of the impact beam, and provide the impact position signal and the vibration signal generated by the impact to the signal processing system.

[0031] The beneficial effects of the present invention are: providing a high-speed train derailment monitoring method, device and system, by arranging the optical fiber sensor in the impact resistance system and connecting the signal processing system, the distributed optical fiber sensor can realize the reception of the real-time impact signal and the positioning of the impact position, and send the vibration signal generated by the impact to the signal processing system after the impact occurs. The impact intensity is predicted based on the vibration intensity prediction model, and the impact intensity is displayed on the local monitoring platform, and is synchronously sent to the disaster monitoring cloud platform, and the relevant departments organize rescue. At the same time, the present invention improves the impact beam, energy absorption block and crash pier, and utilizes the structural characteristics to absorb the impact energy, so that the impact force between the derailed train and the crash pier is greatly reduced, thereby preventing the derailed train from rushing out of the railway structure, such as rushing down the bridge to cause secondary damage to the property and safety under the bridge. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic flow chart of a high-speed train derailment monitoring method described in an embodiment of the present invention;

[0033] Figure 2 Schematic diagram of the operation of the convolutional layer filter described in an embodiment of the present invention;

[0034] Figure 3 This is a schematic structural diagram of the high-speed train derailment monitoring device of the present invention;

[0035] Figure 4 Schematic diagram of the structure of the high-speed train derailment monitoring system described in an embodiment of the present invention;

[0036] Figure 5 A three-dimensional structural diagram of a crash barrier in an embodiment of the present invention;

[0037] Figure 6 is a side view of a crash barrier according to an embodiment of the present invention;

[0038] Figure 7 is a three-dimensional structural diagram of a beam in an embodiment of the present invention;

[0039] Figure 8 A side view of a W-shaped steel according to an embodiment of the present invention;

[0040] Fig. 9It is a three-dimensional structural diagram of the energy absorbing block in an embodiment of the present invention;

[0041] Fig.10 A side view of a crash barrier according to an embodiment of the present invention

[0042] Fig.11 This is a schematic diagram of the structure of the high-speed train derailment monitoring device described in Embodiment 2 of the present invention.

[0043] Markings in the figure:

[0044] 1. Crash barrier; 11. Fiber optic sensor; 12. Holes; 2. Crash pier; 21. Steel bars; 22. Steel pipes; 3. Energy absorbing blocks; 4. Impact beam; 5. Signal processing system; 6. Local monitoring platform; 7. Disaster monitoring cloud platform;

[0045] 800, high-speed train derailment monitoring equipment; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. DETAILED DESCRIPTION

[0046] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the structure of the present invention.

[0047] like Figure 1 In one embodiment of the present invention shown, since there is not enough real data on train impact events for training, earthquake load data similar to impact loads are used as sample data sets to train the model.

[0048] A high-speed train derailment monitoring method comprises the following steps:

[0049] S1: Classifying vibration intensity levels: collecting earthquake motion data of different magnitudes as vibration determination data D1, and classifying the vibration determination data D1 into different vibration intensity levels according to the level of the vibration determination data D1.

[0050] Specifically, S11. Select seismic motion data in the magnitude range of Ms2.0-Ms7.4 from the database; this embodiment selects seismic motion data from January 1, 2022 to December 31, 2022 from the database of the Japan Institute of Disaster Prevention Science and Technology, and the data sampling frequency is 100 Hz.

[0051] S12. Select seismic waves with a recording time of at least 3 seconds and an epicenter within a preset longitude and latitude range from the seismic motion data; in this embodiment, select seismic waves with a time window of 0-3s after the first arrival of the P wave and an epicenter longitude range of 135°E-140°E and an epicenter latitude range of 35°N-39°N.

[0052] S13. The selected seismic waves of different magnitudes are divided into impact waves of corresponding vibration intensities. Specifically, according to the corresponding relationship between the impact wave and the seismic wave, the seismic waves of each magnitude are divided into an impact wave of a vibration intensity.

[0053] Based on the above embodiments, the method further includes:

[0054] S2: Establishing a sample data set: preprocessing the vibration determination data D1, constructing a sample data set D2 from the preprocessed vibration determination data D1, and dividing the sample data set D2 into a training set D3 and a test set D4 in a ratio of 7:3.

[0055] Specifically, the step S2 further includes:

[0056] S21. Automatically pick up seismic waves using long-short time average and Akaike criterion to obtain the arrival time i of the seismic waves in the vertical direction (UD direction);

[0057] S22. Screening and removing abnormal information of the picked-up seismic waves;

[0058] S23. Extract sampling points according to the preset sampling frequency from the eliminated seismic waves to form an acceleration time history curve. Specifically, a total of 600 sampling points from (i-299) to (i+300) are selected for the UD, EW, and NS direction seismic waves to form a total acceleration time history curve of 6s;

[0059] S24. A sample data set of 600*3 is formed by all acceleration time history curves.

[0060] Based on the above embodiments, the method further includes:

[0061] S3: Establish a vibration intensity prediction model P: Build a convolution-long-short time recursive series neural network model, use the training set D3 and the test set D4 to train and test the convolution-long-short time recursive series neural network model, and obtain a trained vibration intensity prediction model P.

[0062] Specifically, Figure 2 As shown, step S3 includes:

[0063] S31. The convolution-long-short time recursive tandem neural network model is composed of an input layer, three convolution layers, three pooling layers, three recursive neural network layers, a fully connected layer, and an output layer. Specifically, the filter size of the convolution layer is 3, and the moving step of the filter is 1. The working diagram is as follows: Figure 2 As shown; the filter and moving step size of the pooling layer are both 3.

[0064] S32. Input the acceleration time history curve and the corresponding vibration intensity in the training set into the convolution-long and short time recursive series neural network model for training;

[0065] Specifically, the step S32 includes:

[0066] S321. Input the acceleration time history curve into the convolution layer to extract feature information;

[0067] S322. Input the extracted feature information into the pooling layer, and use maximum pooling to reduce the dimension of the feature information;

[0068] S323. Input the sampled feature information into the recursive neural network layer, and the recursive neural network layer uses the Sigmoid activation function to extract the time series in the feature information to obtain the memory output information.

[0069] In this embodiment, the recurrent neural network layer can alleviate the gradient vanishing problem existing in the standard recurrent neural network. The recurrent neural network layer uses update gates and reset gates to memorize information:

[0070] Among them, the update gate z t The calculation formula is: t =σ(W (z) x t +U (z) h t-1 )(1), where σ represents the Sigmoid activation function, x t represents the input of the model at time t, W (z) Represents x t The weight matrix, h t-1 represents the hidden layer output at time t-1, U (z) Indicates h t-1 Weight matrix; the update gate can help the model determine how much historical information needs to be brought into the future, z t The closer the elements in are to 1, the more historical information the model will tend to retain, and the less historical information it retains, the more recent input x will be used. t .

[0071] Among them, the reset gate r t The calculation formula is: t =σ(W (r) x t +U (r) h t-1 )(2), where W (r) Represents x t The weight matrix, W (r) Not equal to W (z) ; The reset gate can help the model determine how much historical information should be retained.

[0072] Using the reset gate r t Get the candidate hidden state h′ t , h′ t =tanh(Wx t +r t ⊙Uh t-1 )(3), where ⊙ represents the XOR operation, tanh represents the hyperbolic tangent function, and U represents h t-1 The weight matrix of x t The weight matrix of .

[0073] Finally, the model determines how much information to obtain from the candidate hidden state and how much information to obtain from the hidden state of the previous time step through the update gate, and obtains the final memory output information h t :h t =z t ⊙h t-1 +(1-z t )⊙h′ t (4).

[0074] S324. The fully connected layer performs dimensionality reduction processing on the memory output information to obtain one-dimensional data, and outputs the one-dimensional data as predicted vibration intensity through the output layer.

[0075] S33. Input the acceleration time history curve in the test set into the trained convolution-long-short time recursive series neural network model for filtering and sampling, and output the predicted vibration intensity;

[0076] S34. Calculate the mean square error based on the predicted vibration intensity and the actual vibration intensity in the test set. When the loss function is less than a preset threshold, the convolution-long-short time recursive series neural network model training is completed to obtain a vibration intensity prediction model.

[0077] The method further comprises:

[0078] S4: Predicting the real-time impact intensity level: When an impact occurs, the real-time vibration signal D5 is collected by the optical fiber sensor 11, the real-time vibration signal D5 is preprocessed, and the preprocessed real-time vibration signal D5 is input into the vibration intensity prediction model P, and the vibration intensity prediction model P outputs the predicted vibration intensity.

[0079] Specifically, step S4 includes:

[0080] S41. Obtain vibration signals of more than 3s duration collected by the optical fiber sensor 11. Vibration signals of less than 3s duration may be vibration signals generated by other devices accidentally touching the optical fiber sensor 11, so they are discarded;

[0081] S42. Extracting sampling points in the vibration signal according to a preset sampling frequency to form an acceleration time history curve;

[0082] S43. Input the acceleration time history curve into a vibration intensity prediction model, and the vibration intensity prediction model outputs a predicted vibration intensity, wherein when the vibration intensity corresponding to the acceleration time history curve is less than the vibration intensity in the sample data set, the vibration intensity prediction model outputs a predicted vibration intensity of 0.

[0083] In this embodiment, the vibration intensity collected in the sample data set includes the minimum vibration signal caused by the train impact. Therefore, when the identified vibration intensity is less than the minimum vibration intensity in the sample data set, it can be judged as the vibration caused by the accidental touch of other devices and can be ignored, thereby reducing misjudgment to a certain extent.

[0084] S5: Send prediction results and alarm: send the predicted vibration intensity D6 to the disaster monitoring cloud platform so that the disaster monitoring cloud platform sends an alarm signal.

[0085] To implement this monitoring method, Figure 3 As shown, the present invention provides a high-speed train derailment monitoring device, including an acquisition module, a preprocessing module, a model building module, a prediction module and an alarm module.

[0086] Among them, the acquisition module collects seismic data of different magnitudes as vibration determination data D1, and divides the vibration data into different vibration intensity levels.

[0087] Specifically, the acquisition module includes a first selection unit: selecting seismic motion data in the magnitude range of Ms2.0-Ms7.4 from the database; a second selection unit: selecting seismic waves with a recording time of at least 3 seconds and an epicenter within a preset longitude and latitude range from the seismic motion data; and a division unit: dividing the selected seismic waves of different magnitudes into impact waves of corresponding vibration intensities.

[0088] Wherein, the preprocessing module preprocesses the vibration determination data D1, constructs a sample data set D2 from the preprocessed vibration determination data D1, and divides the sample data set D2 into a training set D3 and a test set D4.

[0089] Specifically, the preprocessing module includes: an automatic picking unit: using long-short time average and Akaike criterion to automatically pick up seismic waves to obtain the arrival time of the seismic waves in the vertical direction; a screening unit: screening the abnormal information of the picked seismic waves and eliminating them; a first sampling unit: extracting sampling points according to a preset sampling frequency in the seismic waves after elimination to form an acceleration time history curve; a data set construction unit: a sample data set is constructed by all the acceleration time history curves.

[0090] Among them, the model building module: builds a convolution-long-short time recursive series neural network model, uses the training set D3 and the test set D4 to train and test the convolution-long-short time recursive series neural network model, and obtains the trained vibration intensity prediction model P.

[0091] Specifically, the model building module includes: a model building unit: a convolution-long-short time recursive series neural network model is composed of an input layer, a convolution layer, a pooling layer, a recursive neural network layer, a fully connected layer and an output layer; a training unit: the acceleration time-history curve and the corresponding vibration intensity in the training set D3 are input into the convolution-long-short time recursive series neural network model for training; a testing unit: the acceleration time-history curve in the test set D4 is input into the trained convolution-long-short time recursive series neural network model for filtering and sampling, and then the predicted vibration intensity is output; a model correction unit: the mean square error is calculated according to the predicted vibration intensity and the actual vibration intensity in the test set D4. When the mean square error is less than a preset threshold, the convolution-long-short time recursive series neural network model training is completed to obtain a vibration intensity prediction model.

[0092] In this embodiment, the training unit includes: a feature information extraction unit: inputting the acceleration time-history curve into the convolution layer to extract feature information; a dimension reduction unit: inputting the extracted feature information into the pooling layer, and using maximum pooling to perform dimension reduction sampling on the feature information; a time series extraction unit: inputting the sampled feature information into the recursive neural network layer, and the recursive neural network layer uses an activation function to extract the time series in the feature information to obtain memory output information; an output unit: the fully connected layer performs dimension reduction processing on the memory output information to obtain one-dimensional data, and outputs the one-dimensional data as predicted vibration intensity through the output layer.

[0093] In this embodiment, the prediction module includes: an acquisition unit: acquiring a vibration signal of more than 3s duration collected by the optical fiber sensor 11; a second sampling unit: extracting sampling points in the vibration signal according to a preset sampling frequency to form an acceleration time history curve; a prediction unit: inputting the acceleration time history curve into a vibration intensity prediction model, and the vibration intensity prediction model outputs a predicted vibration intensity. Among them, when the vibration intensity corresponding to the acceleration time history curve is less than the vibration intensity in the sample data set, the vibration intensity prediction model outputs a predicted vibration intensity of 0.

[0094] Among them, the prediction module: obtains the real-time vibration signal D5 collected by the optical fiber sensor 11 set on the impact beam 4, preprocesses the real-time vibration signal D5, inputs the preprocessed real-time vibration signal D5 into the vibration intensity prediction model P, and the vibration intensity prediction model P outputs the predicted impact intensity D6.

[0095] Among them, the alarm module: sends the impact intensity D6 to the disaster monitoring cloud platform, so that the disaster monitoring cloud platform sends an alarm signal.

[0096] like Figure 4 As shown, in order to realize the monitoring and alarm of high-speed train derailment, the present application also provides a high-speed train derailment monitoring system, including:

[0097] The anti-impact system includes a crash barrier 1 arranged on both sides of the track, the crash barrier 1 includes a transversely arranged impact beam 4, and vertically arranged crash piers 2 are arranged at intervals on the impact beam 4. The impact beam 4 and the crash pier 2 are fixedly connected by an energy absorbing block 3, and a plurality of optical fiber sensors 11 are arranged on the inner side of the impact beam 4.

[0098] The signal processing system 5 is electrically connected to the optical fiber sensor 11, receives the real-time vibration signal D5 collected by the optical fiber sensor 11, and calculates the impact intensity D6 of the real-time vibration signal D5.

[0099] The local monitoring platform 6 is used to receive and display the impact intensity D6 calculated by the signal processing system 5 .

[0100] The disaster monitoring cloud platform 7 is electrically connected to the signal processing system 5 and the optical fiber sensor 11, receives the vibration position collected by the optical fiber sensor 11 and the impact intensity D6 calculated by the signal processing system 5, and sends an alarm signal.

[0101] like Figure 5-8 As shown, the impact beam 4 can absorb 50%-60% of the total energy as the first line of defense against impact. Among them, a W-shaped steel is 6m long, 0.01m thick, and 0.56m in overall cross-section height. The wave height a on both sides is 0.2m, the depth c is 0.17m, and the height b of the middle connection part is 0.16m. Holes 12 are reserved on the impact beam 4 to facilitate connection with the energy absorbing block 3, and a connection point for the optical fiber sensor 11 is set in the inner groove of the impact beam 4 to facilitate the installation of the optical fiber sensor 11.

[0102] The crash piers 2 are located outside the impact beam 4 and are spaced apart along the length direction of the impact beam 4;

[0103] The energy absorbing block 3 is located between the crash pier 2 and the impact beam 4, and is used to fix the impact beam 4 to the crash pier 2. The energy absorbing block 3 can absorb part of the impact energy, further reduce the energy transmitted to the crash pier 2, and reduce the design size of the crash pier 2.

[0104] Specifically, Fig. 9As shown, the energy absorbing block 3 is a hollow structure composed of three curved panels and a rectangular plate. The rectangular plate is fixedly connected to the impact beam 4. The curved panel on the opposite side of the rectangular plate fits the shape of the crash pier 2 and is fixedly connected to the crash pier 2. When the impact load comes, it plays a role in buffering and absorbing energy. Preferably, the energy absorbing block 3 uses steel with a thickness of 0.01m.

[0105] like Fig.10 As shown, the crash pier 2 includes a steel bar 21 and a steel pipe 22 fixed to the outside of the steel bar 21, the steel bar 21 is pre-buried below the ground and extends to all sides, the steel pipe 22 is fixedly connected to the ground, and the gap between the steel pipe 22 and the steel bar 21 is also filled with concrete.

[0106] The outer diameter of the steel pipe 22 is 0.16m and the height is 1m; the radius of the internal concrete column is 0.15m and the height is 1m.

[0107] It should be noted that, regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0108] Embodiment 2: Corresponding to the above method embodiment, this embodiment also provides a high-speed train derailment monitoring device. The high-speed train derailment monitoring device described below and the high-speed train derailment monitoring method described above can refer to each other.

[0109] Fig.11 FIG. 8 is a block diagram of a high-speed train derailment monitoring device 800 according to an exemplary embodiment. Fig.11 As shown, the high-speed train derailment monitoring device 800 may include: a processor 801 and a memory 802. The high-speed train derailment monitoring device 800 may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0110] Wherein, processor 801 is used to control the overall operation of this high-speed train derailment monitoring device 800, to complete all or part of the steps in the above-mentioned high-speed train derailment monitoring method.Memory 802 is used to store various types of data to support the operation of this high-speed train derailment monitoring device 800, these data for example can include the instruction of any application program or method for operating on this high-speed train derailment monitoring device 800, and the data related to application program, for example contact data, message, picture, audio, video etc. that send and receive.

[0111] The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0112] The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone, and the microphone is used to receive external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules may be a keyboard, a mouse, buttons, etc. These buttons may be virtual buttons or physical buttons.

[0113] The communication component 805 is used for wired or wireless communication between the high-speed train derailment monitoring device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0114] Embodiment 3: The high-speed train derailment monitoring device 800 can be implemented by one or more application specific integrated circuits (Application Specific Integrated Circuit, referred to as ASIC), digital signal processors (Digital Signal Processor, referred to as DSP), digital signal processing devices (Digital Signal Processing Device, referred to as DSPD), programmable logic devices (Programmable Logic Device, referred to as PLD), field programmable gate arrays (Field Programmable Gate Array, referred to as FPGA), controllers, microcontrollers, microprocessors or other electronic components, and is used to execute the above-mentioned high-speed train derailment monitoring method.

[0115] Embodiment 4: A computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the high-speed train derailment monitoring method described above are implemented. For example, the computer-readable storage medium can be the memory 802 including the program instructions, and the program instructions can be executed by the processor 801 of the high-speed train derailment monitoring device 800 to complete the high-speed train derailment monitoring method described above.

[0116] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A high-speed train derailment monitoring method, characterized in that: include: S1: Classify vibration intensity levels: collect earthquake motion data of different magnitudes as vibration determination data D1, and classify them into different vibration intensity levels according to the level of vibration determination data D1; S2: Establishing a sample data set: preprocessing the vibration determination data D1, constructing a sample data set D2 from the preprocessed vibration determination data D1, and dividing the sample data set D2 into a training set D3 and a test set D4; S3: Establishing a vibration intensity prediction model P: Building a convolution-long-short time recursive series neural network model, using the training set D3 and the test set D4 to train and test the convolution-long-short time recursive series neural network model, and obtaining a trained vibration intensity prediction model P; S4: predicting the real-time impact intensity level: when the impact occurs, the real-time vibration signal D5 is collected by the optical fiber sensor, the real-time vibration signal D5 is preprocessed, the preprocessed real-time vibration signal D5 is input into the vibration intensity prediction model P, and the vibration intensity prediction model P outputs the predicted vibration intensity D6; S5: Send prediction results and alarm: send the predicted vibration intensity D6 to the disaster monitoring cloud platform so that the disaster monitoring cloud platform sends an alarm signal.

2. A high-speed train derailment monitoring method according to claim 1, characterized in that: In step S1, seismic waves of different magnitudes are selected and divided into impact waves of corresponding intensities; the seismic motion data of different magnitudes are seismic motion data in the magnitude range of Ms2.0-Ms7.4 selected from the database; and seismic waves with a recording time of at least 3 seconds and an epicenter within a preset longitude and latitude range are selected from the seismic motion data.

3. A high-speed train derailment monitoring method according to claim 2, characterized in that: The pre-processing work in steps S2 and S4 includes: The long-short time average and Akaike criterion are used to automatically pick up the seismic waves in the database to obtain the vertical arrival time of the seismic waves; the abnormal information of the picked seismic waves is screened and eliminated; Extracting sampling points from the eliminated seismic waves according to a preset sampling frequency to form an acceleration time history curve; The sample data set D2 is composed of all acceleration time history curves.

4. A high-speed train derailment monitoring method according to claim 3, characterized in that: The convolution-long short time recursive tandem neural network model in step S3 includes an input layer, a convolution layer, a pooling layer, a recursive neural network layer, a fully connected layer and an output layer; The acceleration time history curve and the corresponding vibration intensity in the training set D3 are input into the convolution-long-short time recursive series neural network model for training; The acceleration time history curve in the test set D4 is input into the trained convolution-long-short time recursive series neural network model for filtering and sampling, and the predicted vibration intensity is output; The mean square error is calculated based on the predicted vibration intensity and the actual vibration intensity in the test set D4. When the mean square error is less than a preset threshold, the training of the convolution-long-short time recursive series neural network model is completed, and the vibration intensity prediction model P is obtained.

5. A high-speed train derailment monitoring device, characterized in that: The monitoring device implements a high-speed train derailment monitoring method according to any one of claims 1 to 4, including an acquisition module, a preprocessing module, a model building module, a prediction module and an alarm module, wherein: Acquisition module: collects earthquake motion data of different magnitudes as vibration determination data D1, and divides the vibration determination data D1 into different vibration intensity levels according to the level of the vibration determination data D1; Preprocessing module: preprocessing the vibration determination data D1, constructing a sample data set D2 from the preprocessed vibration determination data D1, and dividing the sample data set D2 into a training set D3 and a test set D4; Model building module: building a convolution-long-short time recursive series neural network model, using the training set D3 and the test set D4 to train and test the convolution-long-short time recursive series neural network model, and obtaining a trained vibration intensity prediction model P; Prediction module: obtains the real-time vibration signal D5 collected by the optical fiber sensor arranged on the impact beam, pre-processes the real-time vibration signal D5, inputs the pre-processed real-time vibration signal D5 into the vibration intensity prediction model P, and the vibration intensity prediction model P outputs the predicted impact intensity D6; Alarm module: sends the impact intensity D6 to the disaster monitoring cloud platform, so that the disaster monitoring cloud platform sends an alarm signal.

6. A high-speed train derailment monitoring device according to claim 5, characterized in that , the acquisition module includes: The first selection unit: selects seismic data in the magnitude range of Ms2.0-Ms7.4 from the database; A second selection unit is used to select earthquake waves with a recording time of at least 3 seconds and an epicenter within a preset longitude and latitude range from the earthquake motion data; Division unit: Divide the selected seismic waves of different magnitudes into impact waves of corresponding vibration intensities.

7. A high-speed train derailment monitoring device according to claim 5, characterized in that ,The preprocessing module includes: Automatic picking unit: Automatically pick up seismic waves using long-short time average and Akaike criterion to obtain the vertical arrival time of seismic waves; Screening unit: Screening and eliminating abnormal information of the picked-up seismic waves; The first sampling unit: extracts sampling points according to a preset sampling frequency from the eliminated seismic waves to form an acceleration time history curve; Data set composition unit: All acceleration time history curves constitute the sample data set D2.

8. A high-speed train derailment monitoring device according to claim 5, characterized in that: The model building module includes: Model building unit: The convolution-long-short time recursive tandem neural network model is composed of input layer, convolution layer, pooling layer, recursive neural network layer, fully connected layer and output layer; Training unit: input the acceleration time history curve and the corresponding vibration intensity in the training set D3 into the convolution-long-short time recursive series neural network model for training; Testing unit: input the acceleration time history curve in the test set D4 into the trained convolution-long-short time recursive series neural network model for filtering and sampling, and then output the predicted vibration intensity; Model correction unit: calculate the mean square error according to the predicted vibration intensity and the actual vibration intensity in the test set D4. When the mean square error is less than the preset threshold, the training of the convolution-long-short time recursive series neural network model is completed to obtain the vibration intensity prediction model P.

9. A high-speed train derailment monitoring system, characterized in that: The monitoring system adopts any monitoring device described in claims 6-8, including: The anti-impact system includes crash barriers arranged on both sides of the track, the crash barriers include a horizontally arranged crash beam and a vertically arranged crash pier, the crash beam and the crash pier are fixedly connected by an energy absorbing block, and a plurality of optical fiber sensors are arranged on the inner side of the crash beam; A signal processing system, the signal processing system is electrically connected to the optical fiber sensor, receives the real-time vibration signal D5 collected by the optical fiber sensor, and processes the real-time vibration signal D5 to obtain the impact intensity D6; The local monitoring platform is used to receive and display the impact intensity D6 calculated by the signal processing system; The disaster monitoring cloud platform is electrically connected to the signal processing system and the optical fiber sensor, receives the vibration position collected by the optical fiber sensor and the impact intensity D6 calculated by the signal processing system, and sends an alarm signal.

10. A high-speed train derailment anti-shock system according to claim 9, characterized in that: The optical fiber sensors are evenly arranged inside the impact beam and provide the impact position signal and the vibration signal generated by the impact to the signal processing system.

Images