CN113095364A - Method, medium, and apparatus for extracting high-speed rail seismic event using convolutional neural network - Google Patents
Method, medium, and apparatus for extracting high-speed rail seismic event using convolutional neural network Download PDFInfo
- Publication number
- CN113095364A CN113095364A CN202110269510.9A CN202110269510A CN113095364A CN 113095364 A CN113095364 A CN 113095364A CN 202110269510 A CN202110269510 A CN 202110269510A CN 113095364 A CN113095364 A CN 113095364A
- Authority
- CN
- China
- Prior art keywords
- speed rail
- training
- samples
- sample
- seismic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 18
- 238000000034 method Methods 0.000 title claims description 38
- 238000012549 training Methods 0.000 claims abstract description 64
- 238000012360 testing method Methods 0.000 claims abstract description 33
- 238000000605 extraction Methods 0.000 claims abstract description 22
- 238000012545 processing Methods 0.000 claims abstract description 22
- 238000010606 normalization Methods 0.000 claims abstract description 7
- 239000013598 vector Substances 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 5
- 230000011218 segmentation Effects 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 10
- 238000004590 computer program Methods 0.000 description 9
- 230000006870 function Effects 0.000 description 7
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000002955 isolation Methods 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a high-speed rail earthquake event extraction method, medium and equipment by utilizing a convolutional neural network, wherein corresponding original earthquake records are subjected to normalization processing according to statistical parameters of all detectors; setting the time sequence length of a single sample according to the time sequence length parameter of the high-speed rail earthquake event; segmenting and intercepting the complete seismic record according to the time sequence length of a single sample, and constructing a training sample set and a test sample set; constructing a 1DCNN network according to the self characteristics of the single sample and setting corresponding hyper-parameters; setting a label for judging whether all samples of the training data set belong to the high-speed rail earthquake event or not; setting network training parameters, and sending the training sample set and the labels into a 1DCNN network for classification training; applying the trained network to the classification and judgment of the actual test data set; and determining the time sequence range of the high-speed rail earthquake event according to the prediction label and the sample serial number of the actual test data set, and completing the extraction of the high-speed rail earthquake event.
Description
Technical Field
The invention belongs to the technical field of exploration geophysics, and particularly relates to a high-speed rail seismic event extraction method, medium and equipment by utilizing a convolutional neural network.
Background
There has been a long history of monitoring and research aimed at train operating conditions and geological environments. In 2004, li et al initially proposed the feasibility of train vibration as a seismic source to detect underground structures. In 2015, the train running state is diagnosed by utilizing the frequency characteristics of the motor train vibration monitoring signals, such as Lizhimin and the like, so that the working condition types of the trains can be distinguished, and the data basis of the train running state is provided. In 2017, a great deal of earthquake records are carried out on the Jingjin intercity railway by Xuohui and the like, which shows that the waveform of the vibration signal can be used as the early warning of the abnormal state of the track and the overhead structure. In 2019, Liu Lei and the like construct a convolutional neural network model taking time-frequency spectrum images corresponding to high-speed rail seismic records as input, train a classifier by artificially marking positive and negative event samples, and finally apply the classifier to screening of a large number of real high-speed rail seismic event records. The future research of applying the high-speed rail earthquake to the geological environment still has great potential, and the accurate and fast acquisition of the real high-speed rail earthquake event record is the first premise for analyzing and processing the earthquake signals in the later period. The existing method for extracting the high-speed rail seismic event record mainly comprises the following steps:
prior art 1: high-speed rail seismic events were obtained by manual interception: the high-speed rail signal is judged by directly visualizing the observation data, and then the corresponding high-speed rail earthquake event is obtained by manual interception, but the high-speed rail earthquake event cannot be automatically extracted. In addition, when the seismic data recording is large, the method needs excessive time and energy, and a large number of high-speed rail seismic events cannot be rapidly acquired.
Prior art 2: a convolutional neural network model taking a time spectrum image as input: firstly, positive and negative event samples of a high-speed rail earthquake record are marked artificially, a convolutional neural network model taking a time-frequency spectrogram image of the sample as input is constructed, and a real high-speed rail earthquake event is screened out by utilizing the model. The method needs to obtain a time-frequency spectrum image of the seismic record, so that when the seismic record data volume is too large, more time is needed to extract a large number of high-speed rail seismic events.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a method, medium and equipment for extracting high-speed rail seismic events by using a convolutional neural network, aiming at the defects in the prior art, wherein a large number of high-speed rail seismic events are extracted by using seismic data acquired by a plurality of detectors outside a high-speed rail line isolation region and a method for training a 1D convolutional neural network model, so that data are provided for subsequent geological information analysis.
The invention adopts the following technical scheme:
the high-speed rail earthquake event extraction method by using the convolutional neural network comprises the following steps of:
s1, carrying out normalization processing on the corresponding original seismic records according to the statistical parameters of all the detectors to obtain normalized complete seismic records;
s2, setting the time sequence length of a single sample according to the time sequence length parameter of the high-speed rail earthquake event;
s3, carrying out segmentation interception on the complete seismic record subjected to normalization processing in the step S1 according to the single sample time sequence length determined in the step S2, and constructing a training sample set and a test sample set;
s4, setting labels for all samples of the training sample set in the step S3 to obtain label vectors of the training sample set;
s5, constructing a 1DCNN network according to the characteristics of the single sample and setting corresponding hyper-parameters;
s6, setting training parameters, and sending the training sample set and the label vector in the step S3 into the 1DCNN constructed in the step S5 for classification training;
s7, applying the network trained in the step S6 to high-speed rail earthquake event classification and judgment of the test sample set to obtain a prediction label;
and S8, determining the time sequence range of the high-speed rail earthquake event according to the prediction label and the sample sequence number obtained in the step S7, and finishing the extraction of the high-speed rail earthquake event.
Specifically, in step S1, it is assumed that the array element number of the detector array is N, and the seismic record of the i-th detector is xi(n) corresponding to a mean value of uiMaximum absolute value of miTo obtain normalized complete seismic record yi(n)：
Wherein i represents the detector number, and takes the value of i as 1,2, …, N.
Specifically, in step S3, the complete seismic record is X ═ X1 … xN]The method comprises E, N, Z components with the largest square difference in three directions, when samples are segmented and intercepted for the complete seismic record, 50% of overlapping length is set between adjacent samples, namely 0.5l of overlapping between the next sample and the previous sample, and the single sample T of the data set is obtained by segmenting and intercepting the seismic recordk(ii) a And then, dividing all samples into a training set and a testing set in proportion, and obtaining the training sample set R and the testing sample set S on the assumption that the segmented samples totally contain K samples.
Specifically, in step S4, it is determined whether all samples contain a high-speed rail earthquake event according to the complete earthquake record waveform, and if yes, the samples belong to the high-speed rail earthquake event, and the label is set to 1; otherwise, the label is set to be 0 to obtain the label vector r of the training data set.
Specifically, in step S5, a 1DCNN network is constructed with 2 convolutional layers + Relu layers, 2 pooling layers, 1 fully-connected layer, and a Softmax layer as structures; the output of the full connection layer is y1,y2,…,ynObtaining output Softmax (y) after Softmax regression processingi(ii) a And setting corresponding super parameters for each layer of the 1DCNN according to the array element number, the sample time sequence length and the time domain waveform characteristic parameters of the seismic record detector.
Specifically, in step S6, the batch processing size and the network learning rate are selected according to the number of training samples, and after Softmax regression processing, the error between the predicted output and the ideal output is calculated, and assuming that p represents the correct probability output and q represents the predicted probability output, the cross entropy loss H (p, q) of the network training is obtained.
Specifically, in step S7, after the model training is completed, the constructed actual test sample set is sent to the network for detecting the high-speed rail earthquake event, so as to obtain a corresponding prediction label vector S:
s＝[s1 s2 … sK-p]
wherein s isjDenotes the j (th)And (3) a prediction label corresponding to each sample, wherein K is the total number of the samples, and p is the number of the samples of the training sample set.
Specifically, in step S8, after the prediction labels of the actual test sample set are obtained, according to the sample numbers of 0 appearing on both sides of the samples whose prediction labels are continuously 1, it is determined that a plurality of samples whose labels are 1 in this range are one complete high-speed rail seismic event, and all high-speed rail seismic events are extracted from the original seismic records according to the time sequence range.
Another aspect of the invention is a computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods described.
Another aspect of the present invention is a computing device, including:
one or more processors, memory, and one or more programs stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods.
Compared with the prior art, the invention has at least the following beneficial effects:
the invention relates to a method for extracting a high-speed rail earthquake event by using a 1D convolutional neural network (1DCNN) model, which can realize the extraction of the earthquake event excited by a high-speed rail train by using a plurality of geophone data outside an isolation area. According to the method, firstly, integral seismic records of a plurality of detectors are normalized, integral seismic records with original relative amplitude difference reserved are obtained, then the time sequence length of a single sample is selected according to the statistic value of the time sequence length of the high-speed rail seismic event, the integral seismic records of the plurality of detectors are segmented and intercepted into a series of samples, a label of whether part of the samples are the high-speed rail seismic event is set for the samples, then a 1DCNN network is built, hyper-parameters are set to achieve feature extraction, and finally event classification is completed through Softmax regression, so that the high-speed rail seismic event can be extracted according to the prediction label and the sample position of actual test data. Compared with the existing high-speed rail earthquake event extraction method, the method can realize automatic extraction of a large number of high-speed rail earthquake events by obtaining the classifier which is trained only by a small number of earthquake record samples.
Furthermore, the time sequence length of a single sample is set according to the time sequence length parameter of the high-speed rail earthquake event, so that the subsequent high-speed rail earthquake event classification error caused by the fact that the time sequence length of the sample is too large or too small can be avoided.
Furthermore, when the multi-channel complete seismic record data is segmented, 50% of overlapping length is set between adjacent samples, so that high-speed rail seismic events can be identified according to classification labels and sample serial numbers of the adjacent samples, and more accurate estimation of the positions of the high-speed rail seismic events can be obtained.
Furthermore, according to the statistical characteristics of the seismic signals in the three directions of the geophone E, N, Z, the seismic record in the direction with the largest variance is selected to construct a sample data set, which is beneficial to reducing the data volume participating in network training and removing the relevant redundant features.
Furthermore, labels of whether the segmented and intercepted part of samples belong to high-speed rail earthquake events are set, and a sample classification basis is provided for the training of a subsequent network model.
Further, in step S5, a 1DCNN network is constructed according to the characteristics of the sample itself and the hyper-parameters are set, so as to provide a mapping model for implementing the sample data to the classification tags, which is beneficial to the subsequent completion of the identification of the high-speed rail seismic event.
Further, based on the actual test sample segmented and intercepted in the step S3, the high-speed rail earthquake event classification detection is performed on the actual test sample, and the sample belonging to the high-speed rail earthquake event is determined, which is beneficial to the subsequent completion of the extraction of the high-speed rail earthquake event.
Further, the location of the high-speed rail seismic event is determined according to the prediction tag and the sample location in the step S7, so that all detected high-speed rail seismic events are extracted accordingly.
In conclusion, the method can effectively and accurately extract the seismic events of the high-speed rail train by using the data of the plurality of geophones, the adopted method is a convolutional neural network deep learning model, and the method has the characteristics of high accuracy, automation, rapidness and the like, and meanwhile, a convenient method for getting rid of manual interception is provided for extracting the seismic events of the high-speed rail train.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a single detector three-component time domain waveform of a single high-speed rail seismic event;
FIG. 3 is a diagram of a 1DCNN high-speed rail seismic event extraction model architecture;
FIG. 4 is a graph of the loss function change during a training session;
fig. 5 is a graph of the results of 30 high-speed rail seismic events.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It will be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It should be further understood that the term "and/or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.
Various structural schematics according to the disclosed embodiments of the invention are shown in the drawings. The figures are not drawn to scale, wherein certain details are exaggerated and possibly omitted for clarity of presentation. The shapes of various regions, layers and their relative sizes and positional relationships shown in the drawings are merely exemplary, and deviations may occur in practice due to manufacturing tolerances or technical limitations, and a person skilled in the art may additionally design regions/layers having different shapes, sizes, relative positions, according to actual needs.
The invention provides a high-speed rail earthquake event extraction method by utilizing a convolutional neural network, which can realize the extraction of a high-speed rail train earthquake event by using a plurality of geophone data outside an isolation region. According to the method, firstly, integral seismic records of a plurality of detectors are normalized, integral seismic records with original relative amplitude difference reserved are obtained, then the time sequence length of a single sample is selected according to the statistic value of the time sequence length of the high-speed rail seismic event, the integral seismic records of the plurality of detectors are segmented and intercepted into a series of samples, a label of whether part of the samples are the high-speed rail seismic event is set for the samples, then a 1DCNN network is built, hyper-parameters are set to achieve feature extraction, and finally event classification is completed through Softmax regression, so that the high-speed rail seismic event can be extracted according to the prediction label and the sample position of actual test data.
Referring to fig. 1, the method for extracting a high-speed rail seismic event by using a convolutional neural network according to the present invention includes the following steps:
s1, normalizing the corresponding original seismic records according to the statistical parameters of the detectors;
assuming that the array element number of the detector array is N, the seismic record of the ith detector is xi(n) corresponding to a mean value of uiMaximum absolute value of miTo obtain its normalized complete seismic record yi(n)：
Wherein i represents the detector number, and takes the value of i as 1,2, …, N.
S2, setting the time sequence length of a single sample according to the time sequence length parameter of the high-speed rail earthquake event;
assuming that the statistical value of the high-speed rail seismic event time sequence length is L, setting the single sample time sequence length L to be 0.6L when constructing the seismic record training and the sample data set to be tested.
S3, intercepting the complete seismic record in a segmented manner according to the time sequence length of the single sample, and constructing a training sample set and a test sample set;
suppose a complete seismic record is X ═ X1 … xN]The method comprises E, N, Z components with the largest square difference in three directions, when samples are segmented and intercepted for the complete seismic record, 50% of overlapping length is set between adjacent samples, namely 0.5l of overlapping between the next sample and the previous sample, and the single sample T of the data set is obtained by segmenting and intercepting the seismic recordk：
Then dividing all samples into a training sample set R and a testing sample set S according to a certain proportion, and assuming that the samples are segmented and intercepted and contain K samples in total, obtaining the training sample set R and the testing sample set S:
R＝[T1 T2 … Tp] (3)
S＝[Tp+1 Tp+2 … TK] (4)
s4, setting a label for judging whether all samples in the training sample set R belong to the high-speed rail earthquake event;
judging whether all samples contain the high-speed rail earthquake event or not according to the complete earthquake record waveform, if so, judging that the samples belong to the high-speed rail earthquake event, and setting a label of '1' for the samples; otherwise, the training data set does not belong to the high-speed rail earthquake event, the label is set to be 0, and the label vector r of the training data set is obtained:
r＝[r1 r2 … rp] (5)
wherein r isjAnd the label corresponding to the jth sample is represented, and the value of the label is '0' or '1'.
S5, constructing a 1DCNN according to the characteristics of the single sample and setting network hyper-parameters;
and aiming at the characteristics of the time sequence data of the sample, constructing a 1DCNN network taking 2 convolutional layers + Relu layers, 2 pooling layers, 1 full connection layer and a Softmax output layer as structures.
Assume the output of the fully-connected layer is y1,y2,…,ynObtaining output Softmax (y) after Softmax regression processingi：
And setting corresponding super parameters for each layer of the 1DCNN according to parameters such as array element number, sample time sequence length, time domain waveform characteristics and the like of the seismic record detector.
S6, setting network training parameters, and sending the training data set and the labels into a 1DCNN network for classification training;
selecting training parameters such as batch processing size and network learning rate according to the number of training samples, calculating an error between prediction output and ideal output after Softmax regression processing, wherein p represents correct probability output, and q represents prediction probability output, so that cross entropy loss H (p, q) of network training is obtained:
s7, applying the trained network to the high-speed rail earthquake event classification judgment of the actual test data set;
after model training is completed, sending the constructed actual test sample set into a network for classification and judgment of the high-speed rail earthquake event to obtain a corresponding prediction label vector s:
s＝[s1 s2 … sK-p] (8)
wherein s isjAnd the prediction label corresponding to the jth sample is represented, and the value of the prediction label is '0' or '1'.
And S8, determining the time sequence range of the high-speed rail earthquake event according to the prediction label and the sample serial number of the test data set, and completing the extraction of the high-speed rail earthquake event.
After the prediction labels of the actual test samples are obtained, according to the sample serial numbers of the samples with the prediction labels continuously being '1' and the samples with the prediction labels continuously appearing '0' on two sides, a plurality of samples with the labels being '1' in the range are judged to be a complete high-speed rail earthquake event, and therefore all the high-speed rail earthquake events can be extracted from the original earthquake records according to the time sequence range of the high-speed rail earthquake event.
In yet another embodiment of the present invention, a terminal device is provided that includes a processor and a memory for storing a computer program comprising program instructions, the processor being configured to execute the program instructions stored by the computer storage medium. The Processor may be a Central Processing Unit (CPU), or may be other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable gate array (FPGA) or other Programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc., which is a computing core and a control core of the terminal, and is adapted to implement one or more instructions, and is specifically adapted to load and execute one or more instructions to implement a corresponding method flow or a corresponding function; the processor of the embodiment of the invention can be used for the operation of the high-speed rail seismic event extraction method by utilizing the convolutional neural network, and comprises the following steps:
carrying out normalization processing on corresponding original seismic records according to the statistical parameters of all the detectors; setting the time sequence length of a single sample according to the time sequence length parameter of the high-speed rail earthquake event; intercepting the normalized complete seismic record in a segmented manner according to the time sequence length of a single sample, and constructing a training sample set and a test sample set; setting a label for judging whether all samples in the training sample set belong to the high-speed rail earthquake event or not; constructing a 1DCNN network according to the self characteristics of the single sample and setting corresponding hyper-parameters; setting network training parameters, and sending the training sample set and the labels into a 1DCNN network for classification training; applying the trained model to the classification and judgment of an actual test sample set; and determining the time sequence range of the high-speed rail earthquake event according to the prediction label and the sample serial number of the actual test sample set, and finishing the extraction of the high-speed rail earthquake event.
In still another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a Memory device in a terminal device and is used for storing programs and data. It is understood that the computer readable storage medium herein may include a built-in storage medium in the terminal device, and may also include an extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space storing an operating system of the terminal. Also, one or more instructions, which may be one or more computer programs (including program code), are stored in the memory space and are adapted to be loaded and executed by the processor. It should be noted that the computer-readable storage medium may be a high-speed RAM memory, or may be a non-volatile memory (non-volatile memory), such as at least one disk memory.
One or more instructions stored in the computer-readable storage medium may be loaded and executed by the processor to implement the corresponding steps of the method for extracting a high-speed rail seismic event using a convolutional neural network in the above embodiments; one or more instructions in the computer-readable storage medium are loaded by the processor and perform the steps of:
carrying out normalization processing on corresponding original seismic records according to the statistical parameters of all the detectors; setting the time sequence length of a single sample according to the time sequence length parameter of the high-speed rail earthquake event; intercepting the normalized complete seismic record in a segmented manner according to the time sequence length of a single sample, and constructing a training sample set and a test sample set; setting a label for judging whether all samples in the training sample set belong to the high-speed rail earthquake event or not; constructing a 1DCNN network according to the self characteristics of the single sample and setting corresponding hyper-parameters; setting network training parameters, and sending the training sample set and the labels into a 1DCNN network for classification training; applying the trained model to the classification and judgment of an actual test sample set; and determining the time sequence range of the high-speed rail earthquake event according to the prediction label and the sample serial number of the actual test sample set, and finishing the extraction of the high-speed rail earthquake event.
Take the signal received by 27 low-frequency detectors arranged at equal intervals along the rail direction at 15m distance from the high-speed rail line when the high-speed rail passes through as an example.
Table 1 shows a sample data set of a high-speed rail seismic record constructed in the present embodiment
Table 2 shows the hyper-parameter settings of the 1DCNN network in this embodiment
| Serial number | Network layer | Dimension of input | Output dimension | Hyper- |
| 1 | Conv1d+Relu | 27×1000 | 16×246 | kernel＝16×(20×27), |
| 2 | MaxPool1d | 16×246 | 16×123 | kernel＝2×1,stride＝2 |
| 3 | Conv1d+Relu | 16×123 | 32×60 | kernel＝32×(5×16), |
| 4 | MaxPool1d | 32×60 | 32×30 | kernel＝2×1,stride＝2 |
| 5 | Full connection | (32×30) | 2 | Is free of |
| 6 | |
2 | 2 | Is free of |
Referring to table 1, table 1 shows a sample data set of a high-speed rail seismic record constructed in this embodiment, where a single sample is a seismic record including 27 receivers and having a time sequence length of 1000, and thus the dimension is 27 × 1000. The training sample set comprises 116 high-speed rail seismic event samples and 3934 non-high-speed rail seismic event samples; the test sample set contained 128 high-speed rail seismic event samples and 3921 non-high-speed rail seismic event samples. Accordingly, the high-speed rail seismic event sample classification label is "1", while the non-high-speed rail seismic event sample classification label is "0", and the detailed division of the data set is detailed in table 1. Please refer to table 2, where table 2 shows the hyper-parameter setting of the 1DCNN network in this embodiment, the seismic records with 27 detectors as input samples and 1000 time sequence lengths, and the network structure includes a convolutional layer + Relu layer, a pooling layer, a full connection layer and a Softmax output layer, and the main input channels, output channels and hyper-parameter setting (convolutional core/pooling core, step length, etc.) of the network are detailed in table 2.
Referring to fig. 2, fig. 2 is a single detector three-component time domain waveform of a single high-speed rail seismic event when a train passes by; the sampling interval was 4ms, for a total of 2008 samples. Referring to fig. 3 and 4, fig. 3 is a structural diagram of a 1DCNN high-speed rail seismic event extraction model, and fig. 4 is a graph showing a loss function variation during a certain training process. It can be seen from fig. 4 that the 1DCNN network has converged through 20 iterations, and thus the time required to complete the model training is short. The data received by the 27 detectors when 30 trains pass through are detected and analyzed by the method, and the 30 high-speed rail earthquake event detection results are shown in fig. 5 and are consistent with the known high-speed rail earthquake event quantity and position of actual earthquake records, so that the high-speed rail earthquake events at the corresponding positions can be sequentially extracted.
In summary, according to the method, the medium and the equipment for extracting the high-speed rail seismic event by using the convolutional neural network, the time domain waveform characteristics of the high-speed rail seismic event are combined, the original one-dimensional seismic record is normalized, segmented and intercepted into a series of single samples, a high-speed rail seismic event data set is constructed, the labels of the training data set are set, the characteristic extraction is realized by training the 1DCNN network, and finally the sample classification is completed through Softmax regression, so that the high-speed rail seismic event can be extracted according to the classification labels and the time sequence range of the samples.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (10)
1. The high-speed rail seismic event extraction method by using the convolutional neural network is characterized by comprising the following steps of:
s1, carrying out normalization processing on the corresponding original seismic records according to the statistical parameters of all the detectors to obtain normalized complete seismic records;
s2, setting the time sequence length of a single sample according to the time sequence length parameter of the high-speed rail earthquake event;
s3, carrying out segmentation interception on the complete seismic record subjected to normalization processing in the step S1 according to the single sample time sequence length determined in the step S2, and constructing a training sample set and a test sample set;
s4, setting labels for all samples of the training sample set in the step S3 to obtain label vectors of the training sample set;
s5, constructing a 1DCNN network according to the characteristics of the single sample and setting corresponding hyper-parameters;
s6, setting training parameters, and sending the training sample set and the label vector in the step S3 into the 1DCNN constructed in the step S5 for classification training;
s7, applying the network trained in the step S6 to high-speed rail earthquake event classification and judgment of the test sample set to obtain a prediction label;
and S8, determining the time sequence range of the high-speed rail earthquake event according to the prediction label and the sample sequence number obtained in the step S7, and finishing the extraction of the high-speed rail earthquake event.
2. The method of claim 1, wherein in step S1, assuming that the array element number of the detector array is N, the seismic record of the ith detector is xi(n) corresponding to a mean value of uiMaximum absolute value of miTo obtain normalized complete seismic record yi(n)：
Wherein i represents the detector number, and takes the value of i as 1,2, …, N.
3. The method of claim 1, wherein in step S3, the complete seismic record is X ═ X1 … xN]The method comprises E, N, Z components with the largest square difference in three directions, when samples are segmented and intercepted for the complete seismic record, 50% of overlapping length is set between adjacent samples, namely 0.5l of overlapping between the next sample and the previous sample, and the single sample T of the data set is obtained by segmenting and intercepting the seismic recordk(ii) a And then, dividing all samples into a training set and a testing set in proportion, and obtaining the training sample set R and the testing sample set S on the assumption that the segmented samples totally contain K samples.
4. The method according to claim 1, wherein in step S4, it is determined whether all samples contain a high-speed rail seismic event according to the complete seismic recording waveform, and if yes, the samples belong to the high-speed rail seismic event, and the label is set to 1; otherwise, the label is set to be 0 to obtain the label vector r of the training data set.
5. The method according to claim 1, wherein in step S5, a 1DCNN network is constructed with 2 convolutional layers + Relu layers, 2 pooling layers, 1 fully-connected layer, and a Softmax layer; the output of the full connection layer is y1,y2,…,ynObtaining output Softmax (y) after Softmax regression processingi(ii) a And setting corresponding super parameters for each layer of the 1DCNN according to the array element number, the sample time sequence length and the time domain waveform characteristic parameters of the seismic record detector.
6. The method of claim 1, wherein in step S6, the batch processing size and the net learning rate are selected according to the number of training samples, and after performing Softmax regression processing, the error between the predicted output and the ideal output is calculated, and assuming that p represents the correct probability output and q represents the predicted probability output, the cross entropy loss H (p, q) of the net training is obtained.
7. The method according to claim 1, wherein in step S7, after the model training is completed, the constructed actual test sample set is sent to the network for detecting the high-speed rail earthquake event, so as to obtain a corresponding predictive label vector S:
s＝[s1 s2 … sK-p]
wherein s isjAnd representing a prediction label corresponding to the jth sample, wherein K is the total number of samples, and p is the number of samples of the training sample set.
8. The method according to claim 1, wherein in step S8, after obtaining the prediction labels of the actual test sample set, according to the sample numbers of 0 appearing on both sides of the samples with the prediction labels of 1 consecutively, several samples with the labels of 1 in the range are determined as a complete high-speed rail seismic event, and all high-speed rail seismic events are extracted from the original seismic record according to the time sequence range.
9. A computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods of claims 1-8.
10. A computing device, comprising:
one or more processors, memory, and one or more programs stored in the memory and configured for execution by the one or more processors, the one or more programs including instructions for performing any of the methods of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110269510.9A CN113095364B (en) | 2021-03-12 | 2021-03-12 | High-speed rail seismic event extraction method, medium and equipment using convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110269510.9A CN113095364B (en) | 2021-03-12 | 2021-03-12 | High-speed rail seismic event extraction method, medium and equipment using convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113095364A true CN113095364A (en) | 2021-07-09 |
| CN113095364B CN113095364B (en) | 2023-12-19 |
Family
ID=76667066
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110269510.9A Active CN113095364B (en) | 2021-03-12 | 2021-03-12 | High-speed rail seismic event extraction method, medium and equipment using convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113095364B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113807305A (en) * | 2021-09-27 | 2021-12-17 | 宫雪峰 | Road destructive prediction automatic driving method based on naive Bayes classification algorithm |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1649289A (en) * | 2004-03-17 | 2005-08-03 | 威盛电子股份有限公司 | Differential phase detecting method and system |
| US20190050711A1 (en) * | 2017-08-08 | 2019-02-14 | Neusoft Corporation | Method, storage medium and electronic device for detecting vehicle crashes |
| WO2019100724A1 (en) * | 2017-11-24 | 2019-05-31 | 华为技术有限公司 | Method and device for training multi-label classification model |
| CN111126471A (en) * | 2019-12-18 | 2020-05-08 | 中国石油大学(华东) | Microseism event detection method and system |
| CN111367961A (en) * | 2020-02-27 | 2020-07-03 | 西安交通大学 | Time sequence data event prediction method and system based on graph convolution neural network and application thereof |
| CN111505710A (en) * | 2020-05-06 | 2020-08-07 | 西安交通大学 | High-speed rail seismic source seismic signal extraction method based on equidistant spectrum characteristics |
-
2021
- 2021-03-12 CN CN202110269510.9A patent/CN113095364B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1649289A (en) * | 2004-03-17 | 2005-08-03 | 威盛电子股份有限公司 | Differential phase detecting method and system |
| US20050207517A1 (en) * | 2004-03-17 | 2005-09-22 | Chunlan Qin | Delta-phase detection method and system |
| US20190050711A1 (en) * | 2017-08-08 | 2019-02-14 | Neusoft Corporation | Method, storage medium and electronic device for detecting vehicle crashes |
| WO2019100724A1 (en) * | 2017-11-24 | 2019-05-31 | 华为技术有限公司 | Method and device for training multi-label classification model |
| CN111126471A (en) * | 2019-12-18 | 2020-05-08 | 中国石油大学(华东) | Microseism event detection method and system |
| CN111367961A (en) * | 2020-02-27 | 2020-07-03 | 西安交通大学 | Time sequence data event prediction method and system based on graph convolution neural network and application thereof |
| CN111505710A (en) * | 2020-05-06 | 2020-08-07 | 西安交通大学 | High-speed rail seismic source seismic signal extraction method based on equidistant spectrum characteristics |
Non-Patent Citations (3)
| Title |
|---|
| DAWEI LIU ETAL.: "Poststack Seismic Data Denoising Based on 3-D Convolutional Neural Network", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》, vol. 58, no. 3, XP011774144, DOI: 10.1109/TGRS.2019.2947149 * |
| 刘磊等: "大量高铁地震事件的属性体提取与特性分析", 《地球物理学报》 * |
| 刘磊等: "大量高铁地震事件的属性体提取与特性分析", 《地球物理学报》, vol. 62, no. 6, 30 June 2019 (2019-06-30), pages 1 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113807305A (en) * | 2021-09-27 | 2021-12-17 | 宫雪峰 | Road destructive prediction automatic driving method based on naive Bayes classification algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113095364B (en) | 2023-12-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11714397B2 (en) | System and method for generating machine learning model with trace data | |
| US11586913B2 (en) | Power equipment fault detecting and positioning method of artificial intelligence inference fusion | |
| CN111797890A (en) | Method and system for detecting defects of power transmission line equipment | |
| CN108711148B (en) | Tire defect intelligent detection method based on deep learning | |
| KR20160130422A (en) | Method and system for checking goods | |
| CN110647830B (en) | Bearing fault diagnosis method based on convolutional neural network and Gaussian mixture model | |
| CN112508105B (en) | Fault detection and retrieval method for oil extraction machine | |
| CN111855810B (en) | Rail foot damage identification method and system based on recurrent neural network | |
| CN101614787A (en) | Analogical Electronics method for diagnosing faults based on M-ary textural classification device | |
| CN113763312A (en) | Detecting defects in semiconductor samples using weak marks | |
| CN106778908A (en) | A kind of novelty detection method and apparatus | |
| CN111783616B (en) | Nondestructive testing method based on data-driven self-learning | |
| CN111753877B (en) | Product quality detection method based on deep neural network migration learning | |
| CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
| CN116012681A (en) | Method and system for diagnosing motor faults of pipeline robot based on sound vibration signal fusion | |
| CN113095364B (en) | High-speed rail seismic event extraction method, medium and equipment using convolutional neural network | |
| CN116466408B (en) | Artificial neural network superbedrock identification method based on aeromagnetic data | |
| CN112147221B (en) | Steel rail screw hole crack identification method and system based on ultrasonic flaw detector data | |
| CN112287595B (en) | Prediction method of post-grouting thickness of shield tunnel wall based on ground penetrating radar detection and machine learning | |
| CN106951924A (en) | Acceleration Algorithm in Seismic Coherence Cube image slices automatic identifying method and system based on AdaBoost algorithms | |
| CN116823793A (en) | Device defect detection method, device, electronic device and readable storage medium | |
| CN112801073B (en) | Acoustic emission first-break signal classification and identification method based on deep neural network | |
| CN112215307B (en) | Method for automatically detecting signal abnormality of earthquake instrument by machine learning | |
| CN114462127A (en) | Structural damage identification method and system based on deep extreme learning machine | |
| CN111860258A (en) | Examination room global event detection method and system based on three-dimensional convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |