Rock mass microseismic identification method, electronic device and storage medium

Abstract

The present invention is applicable to the technical field of microseismic monitoring and early warning, and provides a method for identifying rock mass microseisms, an electronic device, and a storage medium. The method includes: obtaining the original microseismic monitoring waveform and segmenting it to obtain multiple segments of waveforms; for each segment of waveform, extracting time domain features, amplitude distribution statistical features, and time-frequency feature images; inputting the time domain features and amplitude distribution statistical features into the first classification model in a pre-trained dual convolutional neural network model to obtain a first classification result; splicing the time-frequency feature images and inputting them into the second classification model in the dual convolutional neural network model to obtain a second classification result; and determining the microseismic event type based on the first classification result and the second classification result. The present invention can improve the accuracy and efficiency of identifying rock mass microseismic waveform information.

Description

Technical Field

[0001] The present invention belongs to the technical field of microseismic monitoring and early warning, and particularly relates to a method for identifying rock mass microseisms, an electronic device, and a storage medium. Background Art

[0002] During the processes of tunnel excavation, mine exploitation, and underground engineering construction, construction disturbances will cause redistribution of formation stress and local stress concentration. The engineering rock mass will also be accompanied by fracture damage, resulting in an increased risk of instability of the engineering structure and possibly inducing accidents such as collapses and landslides. During the fracture damage process of rock mass engineering materials, stress waves will be released inside the rock mass. By using microseismic monitoring technology to monitor and record the stress waves of the rock mass during construction and then interpreting the stress waveform characteristics, the stability evaluation and prediction of the engineering rock mass can be realized. However, the current identification of rock mass microseismic waveform information is still not accurate and efficient enough. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a method for identifying rock mass microseisms, an electronic device, and a storage medium to improve the accuracy and efficiency of identifying rock mass microseismic waveform information.

[0004] The first aspect of the embodiments of the present invention provides a method for identifying rock mass microseisms, including:

[0005] Obtain the original microseismic monitoring waveform and segment it to obtain multiple segments of waveforms;

[0006] For each segment of waveform, extract time-domain features, amplitude distribution statistical features, and time-frequency feature images;

[0007] Input the time-domain features and amplitude distribution statistical features into the first classification model in a pre-trained dual convolutional neural network model to obtain a first classification result;

[0008] Input the spliced time-frequency feature images into the second classification model in the dual convolutional neural network model to obtain a second classification result;

[0009] Based on the first classification result and the second classification result, determine the microseismic event type.

[0010] Combined with the first aspect, in a possible implementation manner of the first aspect, the time-domain features include at least one of the following:

[0011] Event effective duration, maximum amplitude value, minimum amplitude value, and average amplitude value;

[0012] Among them, the event effective duration is calculated by the following method:

[0013] Calculate the short-time average to long-time average ratio data of this segment of waveform;

[0014] According to Determine the first arrival threshold, and according to determine the termination threshold; wherein, a , b are preset coefficients and are both less than 1; is the difference between the maximum value and the minimum value of the long-term and short-term average ratio; Z is the mode in the long-term and short-term average ratio;

[0015] Determine the first arrival time when the long-term and short-term average ratio is first greater than the first arrival threshold, and determine the termination time when the long-term and short-term average ratio is first less than the termination threshold since the first arrival time;

[0016] Based on the first arrival time and the termination time, determine the effective duration of the event.

[0017] Combined with the first aspect, in a possible implementation manner of the first aspect, the time-frequency feature image includes at least one of the following:

[0018] Fast Fourier transform spectrum diagram, short-time Fourier transform time-frequency diagram plotted with an adaptive color scale, short-time Fourier transform time-frequency diagram plotted with a unified scale color scale, wavelet transform time-frequency diagram plotted with an adaptive color scale, and wavelet transform time-frequency diagram plotted with a unified scale color scale.

[0019] Combined with the first aspect, in a possible implementation manner of the first aspect, extract the amplitude distribution statistical features, including:

[0020] Calculate the number of sampling points corresponding to different amplitudes in this section of the waveform, fit to obtain the amplitude distribution curve, and calculate the amplitude distribution statistical features according to the amplitude distribution curve;

[0021] Among them, the amplitude distribution statistical features include at least one of the following:

[0022] Maximum value, variance, standard deviation, root mean square, root mean square amplitude, kurtosis, skewness, waveform factor, peak factor, impulse factor, margin factor.

[0023] Combined with the first aspect, in a possible implementation manner of the first aspect, both the first classification result and the second classification result are the probabilities of various microseismic event types;

[0024] Based on the first classification result and the second classification result, determine the microseismic event type, including:

[0025] For each microseismic event type, perform weighted summation on the probability in the first classification result and the probability in the second classification result to obtain the comprehensive probability of this microseismic event type;

[0026] Determine the microseismic event type with the largest comprehensive probability as the microseismic event type of this section of the waveform.

[0027] In combination with the first aspect, in a possible implementation of the first aspect, the dual convolutional neural network model is trained in the following manner:

[0028] Obtain different segments of waveforms and divide them into a training set and a test set;

[0029] Train and test the first convolutional neural network model based on the time-domain features, amplitude distribution statistical features of different segments of waveforms in the training set and the test set, and the predetermined microseismic event types to obtain the first classification model;

[0030] Train and test the second convolutional neural network model based on the time-frequency feature images of different segments of waveforms in the training set and the test set to obtain the second classification model;

[0031] Determine the weights of the first classification model and the second classification model to obtain the dual convolutional neural network model to be verified;

[0032] Judge whether the recognition accuracy rate of the dual convolutional neural network model to be verified for the test set reaches the preset standard. If it reaches, determine it as the dual convolutional neural network model.

[0033] In combination with the first aspect, in a possible implementation of the first aspect, training and testing the first convolutional neural network model or the second convolutional neural network model includes:

[0034] After training the target convolutional neural network model, determine the waveform type of each segment of waveform according to the classification result of each segment of waveform in the test set output by the target convolutional neural network model; where the target convolutional neural network model is the first convolutional neural network model or the second convolutional neural network model, and the waveform types include typical waveforms and atypical waveforms;

[0035] Calculate the recognition accuracy rates of the target convolutional neural network model for typical waveforms and atypical waveforms respectively. If the recognition accuracy rates of the target convolutional neural network model for typical waveforms and atypical waveforms are both greater than the preset accuracy rate threshold, determine that the target convolutional neural network model training is completed;

[0036] If the classification accuracy rate of the target convolutional neural network model for typical waveforms and / or atypical waveforms is less than or equal to the accuracy rate threshold, adjust the parameters, training set or test set of the target convolutional neural network model, and retrain or test the target convolutional neural network model.

[0037] In combination with the first aspect, in a possible implementation of the first aspect, the classification result is the probability of various microseismic event types;

[0038] Determining the waveform type of each segment of waveform according to the classification result of each segment of waveform in the test set output by the target convolutional neural network model includes:

[0039] Sort various microseismic event types according to the probability from large to small;

[0040] Calculate the variance of the probabilities of the first two microseismic event types in the sequence to obtain the first variance;

[0041] Calculate the variance of the probabilities of all microseismic event types in the sequence to obtain the second variance;

[0042] If the first variance is greater than or equal to the second variance, determine that this section of waveform is a typical waveform;

[0043] If the first variance is less than the second variance, determine that this section of waveform is an atypical waveform.

[0044] The second aspect of the embodiments of the present invention provides a rock mass microseismic identification device, including:

[0045] An acquisition module, configured to acquire the original microseismic monitoring waveform and segment it to obtain multiple segments of waveforms;

[0046] A feature extraction module, configured to extract time-domain features, amplitude distribution statistical features, and time-frequency feature images for each segment of waveform;

[0047] An identification module, configured to input the time-domain features and amplitude distribution statistical features into the first classification model in the pre-trained dual convolutional neural network model to obtain a first classification result; splice the time-frequency feature images and input them into the second classification model in the dual convolutional neural network model to obtain a second classification result; determine the microseismic event type based on the first classification result and the second classification result.

[0048] The third aspect of the embodiments of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method in the first aspect or any one implementation manner of the first aspect as described above.

[0049] The fourth aspect of the embodiments of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the method in the first aspect or any one implementation manner of the first aspect as described above.

[0050] The beneficial effects of the embodiments of the present invention compared with the prior art are:

[0051] In the embodiments of the present invention, by extracting the time-domain features, amplitude distribution statistical features, and time-frequency feature images of each waveform segment, a waveform feature library is constructed. The quantization parameter features of the waveform, such as time-domain features and amplitude distribution statistical features, are classified through a first classification model, and the time-frequency feature images of the waveform are classified through a second classification model. The classification results of the two models are comprehensively used to identify the types of microseismic events. This solution adopts a large number of features, can accurately identify the types of microseismic events, does not require pre-filtering of the original microseismic monitoring waveforms, avoids the distortion of the original waveforms and the loss of some effective information. And it can realize the automatic identification of common microseismic event types, improving the identification efficiency of a large number of microseismic monitoring waveforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0053] Figure 1 It is a schematic flowchart of the implementation of the rock mass microseismic identification method provided by the embodiments of the present invention;

[0054] Figure 2 It is a schematic diagram of the traditional STA / LTA method provided by the embodiments of the present invention Figure 1 ;

[0055] Figure 3 It is a schematic diagram of the traditional STA / LTA method provided by the embodiments of the present invention Figure 2 ;

[0056] Figure 4 It is a schematic diagram of the improved STA / LTA method provided by the embodiments of the present invention Figure 1 ;

[0057] Figure 5 It is a schematic diagram of the improved STA / LTA method provided by the embodiments of the present invention Figure 2 ;

[0058] Figure 6 It is the training process of the dual convolutional neural network model provided by the embodiments of the present invention Figure 1 ;

[0059] Figure 7 It is the training process of the dual convolutional neural network model provided by the embodiments of the present invention Figure 2 ;

[0060] Figure 8 It is the training process of the dual convolutional neural network model provided by the embodiments of the present invention Figure 3 ;

[0061] Figure 9 It is the training process of the dual convolutional neural network model provided by the embodiments of the present invention Figure 4 ;

[0062] Figure 10 It is a schematic diagram of the microseismic monitoring waveform provided by the embodiments of the present invention;

[0063] Figure 11 It is a schematic diagram of the amplitude distribution curve provided by the embodiments of the present invention;

[0064] Figure 12 It is a schematic diagram of the spliced feature image provided by the embodiments of the present invention;

[0065] Figure 13 It is a schematic structural diagram of the rock mass microseismic identification device provided by the embodiments of the present invention;

[0066] Figure 14 It is a schematic structural diagram of the electronic device provided by the embodiments of the present invention. Detailed implementation manners

[0067] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present invention. However, those skilled in the art should clearly understand that the present invention can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from hindering the description of the present invention.

[0068] In order to illustrate the technical solutions described in the present invention, the following will be described through specific embodiments.

[0069] Currently, there are still many difficulties in the interpretation of microseismic monitoring waveforms. For example, the screening of effective stress waves mainly relies on human experience, and a complete, efficient, and accurate automated waveform classification and recognition method has not yet been formed. Moreover, waveform recognition is performed after pre-data filtering, which is likely to cause distortion of the original waveform and loss of some effective information. Therefore, the embodiments of the present invention propose an automated rock mass microseismic identification method that does not require pre-filtering processing, has a simple process, and high accuracy.

[0070] Figure 1 It is a schematic diagram of the implementation process of the rock mass microseismic identification method proposed by the embodiments of the present invention. Refer to Figure 1 shown. The method includes:

[0071] Step S101, obtain the original microseismic monitoring waveform and segment it to obtain multiple segments of waveforms.

[0072] The original microseismic monitoring waveforms can be obtained using microseismic monitoring equipment, which are in the form of curves with time as the abscissa and amplitude values as the ordinate, denoted as i the amplitude value corresponding to the x i . In this embodiment, the original microseismic monitoring waveforms are segmented with each 1.5 s as a waveform segment. The specific segmentation method can be set according to actual needs. For example, it can also be each 2 s as a waveform segment, or the length of each waveform segment can also be different. This embodiment does not limit this.

[0073] Step S102: For each waveform segment, extract time-domain features, amplitude distribution statistical features, and time-frequency feature images.

[0074] In this embodiment, the signal features of the waveforms are extracted to establish an over-complete feature dictionary. It is mainly divided into three aspects: the time-domain features of the waveforms, amplitude distribution statistical features, and time-frequency feature images. Considering the over-complete theory, when it is not clear which features are sufficient to distinguish the waveforms, the more types of features included, the better.

[0075] (1) Time-domain features

[0076] In this embodiment, the time-domain features may include but are not limited to the effective duration of the event, maximum amplitude value, minimum amplitude value, and average amplitude value, etc.

[0077] The maximum amplitude value is denoted as .

[0078] The minimum amplitude value is denoted as .

[0079] The average amplitude value is denoted as .

[0080] The "effective duration of the event" is different from the length of the waveform (1.5 s). It refers to the time length from the triggering to the termination of the event and can be carried out using the improved short-term average / long-term average ratio method (STA / LTA). The short-term average / long-term average ratio method is to give a sliding long-time window, and then take a short-time window within this window. The endpoints or starting points of the two windows coincide, and the ratio of the short-time window signal average STA to the long-time window signal average LTA is used to reflect the change in signal amplitude or energy. STA mainly reflects the average value of microseismic signals, and LTA mainly reflects the average value of background noise. At the time when the microseismic signal arrives, STA changes faster than LTA, and the corresponding STA / LTA value will have an obvious increase. When its ratio is greater than a certain threshold, it can be determined that a microseismic event has occurred, so as to achieve the purpose of automatically detecting and picking up the microseismic first arrival.

[0081] The basic calculation formula of the short-term average / long-term average ratio method is:

[0082] ;

[0083] In the formula, is the length of the long window; is the length of the short window; is the length by which the time window moves forward; is the current position where the window is located; is the set trigger threshold.

[0084] On this basis, this embodiment proposes a dynamic threshold setting scheme based on the mode of STA / LTA values, including:

[0085] Calculating the short-term and long-term average ratio data of this section of waveform;

[0086] According to determine the first arrival threshold, and according to determine the termination threshold; where a , b are preset coefficients and are both less than 1; is the difference between the maximum and minimum values of the short-term and long-term average ratio; Z is the mode in the short-term and long-term average ratio;

[0087] Determine the first arrival time when the short-term and long-term average ratio is first greater than the first arrival threshold, and determine the termination time when the short-term and long-term average ratio is first less than the termination threshold since the first arrival time;

[0088] Based on the first arrival time and the termination time, determine the effective duration of the event.

[0089] Here, the mode is the number that appears most frequently among all STA / LTA values calculated from a section of waveform. The essence of the STA / LTA method is to find the points where the waveform changes suddenly. In order to exclude the interference of noise on point selection, a threshold needs to be set. However, since this embodiment aims to achieve automated processing of batch waveforms, it is obviously unreasonable to manually set the threshold for the different bottom noises of each waveform; and if a unified "fixed threshold" is used without considering the differences in the bottom noises of each section of waveform, the calculation results have a large error. Using the mode as the baseline of the threshold can achieve different thresholds for each waveform.

[0090] Refer to Figures 2 - 5 as shown, to illustrate the effect of the improved STA / LTA method in this embodiment.

[0091] Under the traditional STA / LTA method, the length of the long window is set to 1.5, the length of the short window is set to 0.14, the first arrival threshold is 0.01, and the termination threshold is 0.005. Under the fixed threshold, for Figure 2 as shown, the recognition effect of the first arrival time and the termination time of waveform 1 is acceptable, but the same first arrival threshold and termination threshold are applied to Figure 3There are very obvious errors when the waveform 2 shown is present, so that there are no intersection points between the STA / LTA ratio curve and the set first arrival threshold or termination threshold. If more accurate identification of waveform 2 is required, it is necessary to manually adjust the first arrival threshold and the termination threshold. This is not conducive to the automated processing of a large number of waveforms.

[0092] Under the improved STA / LTA method of the present invention, the length of the long window is set to 1.5, the length of the short window is set to 0.14, the first arrival threshold coefficient is 0.1, and the termination threshold coefficient is 0.026. Since it is applied to waveform 1 and waveform 2, it will be adjusted respectively based on the mode of the STA / LTA ratio of the two waveforms. Therefore, as Figure 4 and Figure 5 shown, good recognition effects can be obtained in both waveforms.

[0093] (2) Time-frequency characteristic image

[0094] In this embodiment, the time-frequency characteristic image may include but is not limited to visual images obtained by time-frequency analysis techniques such as fast Fourier transform spectrogram, short-time Fourier transform time-frequency diagram, wavelet transform time-frequency diagram, etc.

[0095] In addition, considering that this embodiment is based on the processing of a large number of waveforms, there are differences in the frequency intensity and power spectral density of different waveforms in magnitude. When using an adaptive color scale to draw a cloud map, the details of each image are very rich. However, when the power spectral density corresponding to yellow in one image is 1.2, the same color may correspond to 0.08 in another image. This will lead to errors in the neural network learning and training. Therefore, the step of drawing a cloud map with a unified scale color scale is supplemented.

[0096] That is, when performing time-frequency analysis on each waveform, in order to ensure the accuracy of determining the frequency intensity or power spectral density through color during the image recognition process, first draw a cloud map with an adaptive color scale, and then draw a cloud map with a unified scale color scale. In this embodiment, the fast Fourier transform spectrogram, the short-time Fourier transform time-frequency diagram drawn with an adaptive color scale, the short-time Fourier transform time-frequency diagram drawn with a unified scale color scale, the wavelet transform time-frequency diagram drawn with an adaptive color scale, and the wavelet transform time-frequency diagram drawn with a unified scale color scale are mainly selected as the time-frequency characteristic images.

[0097] (3) Amplitude distribution statistical characteristics

[0098] As a possible implementation, extracting the amplitude distribution statistical characteristics includes: calculating the number of sampling points corresponding to different amplitudes in this waveform segment, fitting to obtain an amplitude distribution curve, and calculating amplitude distribution statistical characteristics according to the amplitude distribution curve.

[0099] For a section of waveform, the range of the vertical coordinate from the minimum amplitude value to the maximum amplitude value is called the amplitude interval, and the number of points on the waveform with the same amplitude value on the vertical coordinate is called the number of sampling points corresponding to that amplitude value. The curve plotted with the amplitude value as the abscissa and the number of sampling points as the ordinate is called the amplitude distribution curve of this section of waveform. In an ideal state, the original waveform is a continuous curve, and its amplitude distribution curve is also a continuous curve. Within the amplitude interval, the number of sampling points corresponding to each amplitude value is always greater than or equal to one. In the actual state, the microseismic waveforms obtained by the microseismic monitoring equipment are all composed of discrete sampling points. The amplitude interval is equally spaced into N small intervals, and the number of sampling points falling within each small interval is counted. Due to technical limitations, when N is extremely large (i.e., the interval length is extremely small), it is very likely that the number of sampling points within the small interval range is 0. This type of data belongs to abnormal data. At this time, it is necessary to obtain the corresponding number of sampling points through interpolation fitting, or appropriately adjust the value of N according to the sampling rate until there is no abnormal data. Taking the median value of the small interval amplitude as the abscissa and the number of sampling points as the ordinate, discrete points are obtained, and the amplitude distribution curve of the waveform is fitted and drawn.

[0100] Based on the amplitude distribution curve for analysis and calculation of the statistical characteristics of the amplitude distribution, compared with directly calculating the above statistical characteristics for the original waveform, after processing based on the amplitude distribution curve, the statistical characteristics of the two types of events have more obvious differences, which helps to improve the classification effect.

[0101] The statistical characteristics of the amplitude distribution include but are not limited to:

[0102] Maximum value, variance, standard deviation, root mean square, root mean square amplitude, kurtosis, skewness, waveform factor, peak factor, impulse factor, margin factor.

[0103] Assume that the total number of intervals divided when forming the amplitude distribution curve is N , the i th interval corresponds to the number of sampling points y i , then the calculation formulas for the above characteristic values are as follows:

[0104] Maximum value: .

[0105] Variance: .

[0106] Standard deviation: .

[0107] Root mean square: .

[0108] Root mean square amplitude: .

[0109] Kurtosis: .

[0110] Skewness: 。

[0111] Form factor: 。

[0112] Crest factor: 。

[0113] Pulse factor: 。

[0114] Margin factor: 。

[0115] Step S103: Input the time-domain features and amplitude distribution statistical features into the first classification model in the pre-trained dual convolutional neural network model to obtain the first classification result.

[0116] In this embodiment, the time-domain features and amplitude distribution statistical features are quantization parameter features, which are aggregated and tiled into a one-dimensional array and input into the first classification model. The first classification model can be a convolutional neural network model, and the model output uses soft labels, that is, it does not directly output the recognition result of a certain waveform segment, but outputs the probability that the waveform segment is various events.

[0117] Step S104: Input the spliced time-frequency feature images into the second classification model in the dual convolutional neural network model to obtain the second classification result.

[0118] In this embodiment, each time-frequency feature image can be tiled to form a feature image of the original microseismic monitoring waveform in the upper left, the fast Fourier transform spectrum in the upper middle, the short-time Fourier transform time-frequency diagram (adaptive color scale) in the upper right, the short-time Fourier transform time-frequency diagram (uniform scale color scale) in the middle right, the wavelet transform time-frequency diagram (adaptive color scale) in the lower left, and the wavelet transform time-frequency diagram (uniform scale color scale) in the lower right, and input it into the second classification model. Similarly, the second classification model can be a convolutional neural network model, and the model output uses soft labels, that is, it does not directly output the recognition result of a certain waveform segment, but outputs the probability that the waveform segment is various events.

[0119] Step S105: Determine the microseismic event type based on the first classification result and the second classification result.

[0120] For each microseismic event type, the probabilities in the first classification result and the second classification result are weighted and summed to obtain the comprehensive probability of the microseismic event type, and the microseismic event type with the largest comprehensive probability is determined as the microseismic event type of the waveform segment.

[0121] In the embodiments of the present invention, by extracting the time-domain features, amplitude distribution statistical features, and time-frequency feature images of each waveform segment, a waveform feature library is constructed. The quantization parameter features of the waveform, such as time-domain features and amplitude distribution statistical features, are classified through a first classification model, and the time-frequency feature images of the waveform are classified through a second classification model. The classification results of the two models are comprehensively used to identify the types of microseismic events. This solution uses a large number of features, can accurately identify the types of microseismic events, does not require pre-filtering of the original microseismic monitoring waveforms, avoids the distortion of the original waveforms and the loss of some effective information. And it can realize the automatic identification of common microseismic event types, improving the identification efficiency of a large number of microseismic monitoring waveforms.

[0122] The dual convolutional neural network model of the above embodiment is obtained through pre-training. In one embodiment, the training process of the dual convolutional neural network model can be referred to Figure 6 as shown

[0123] Step 1: Obtain a plurality of original microseismic monitoring waveforms and segment them to obtain different waveform segments.

[0124] In this embodiment, the segmentation method can be referred to Figure 1 the description in the embodiment.

[0125] Step 2: For each waveform segment, extract the time-domain features, amplitude distribution statistical features, and time-frequency feature images. The specific implementation method can be referred to Figure 1 the description in the embodiment.

[0126] Step 3: Combine the time-domain features, time-frequency feature images, and amplitude distribution statistical features of each waveform segment to obtain a waveform feature library.

[0127] Step 4: Based on the actual situation and existing classification methods, pre-classify the waveform segments involved in the waveform feature library and add category labels to form a dataset for constructing the dual convolutional neural network model. The sample size of each type of event is preferably equal. Here, the types of microseismic events include but are not limited to blasting events, instability and creep, rock-breaking construction disturbance, rock mass fracture microseismic events, invalid noise, or undefined waveforms, etc.

[0128] Step 5: Divide the dataset into a training set and a test set, and further divide them into a quantization database and an image database. That is, form: a training set quantization database, a training set image database, a test set quantization database, and a test set image database.

[0129] Taking the total number of waveforms as 2500 segments as an example, after mixing and scrambling all the waveform segments, 70% (1750 waveform segments) are extracted as the training set, and 30% (750 waveform segments) are used as the test set.

[0130] Step 6:

[0131] See Figure 7 As shown, taking the waveform training set quantization database as the input, the input form is [15×1×1]; preferably: the convolutional neural network structure includes one input layer, three convolutional layers, three ReLU activation layers, three pooling layers, two fully connected layers, and one output layer. The convolutional kernel size is [2×1], the stride is 1, the initial learning rate is set to 0.001, the maximum number of training times is 1000 rounds, the adjustment factor is set to 0.1, and the training algorithm is selected as the gradient descent method. In order to improve the interpretability of the convolutional neural network model, an attention mechanism is incorporated into the first convolutional layer and the last convolutional layer, enabling the network to allocate weights according to the contribution degree of different features to classification, and visualizing the weights in the last convolutional layer. After the loss rate is infinitely close to 0 or the maximum number of training times is reached, the test set quantization feature library is input into the obtained convolutional neural network model to verify the accuracy rate; when the accuracy rate reaches the standard, a first classification model is established. Otherwise, adjust the parameters and retrain until the accuracy rate reaches the standard.

[0132] See Figure 7 As shown, the feature images in the training set image feature library assign values to each pixel point and serve as the input layer; the convolutional neural network structure includes one input layer, three convolutional layers, three ReLU activation layers, three pooling layers, two fully connected layers, and one output layer. The convolutional kernel size is [3×3], the stride is 1, the padding is set to "same", the initial learning rate is set to 0.001, the maximum number of training times is 1000 rounds, the adjustment factor is set to 0.1, and the training algorithm is selected as the gradient descent method. In order to improve the interpretability of the convolutional neural network model, an attention mechanism is incorporated into the first convolutional layer and the last convolutional layer, enabling the network to allocate weights according to the contribution degree of different features to classification, and visualizing the weights in the last convolutional layer. After the loss rate is infinitely close to 0 or the maximum number of training times is reached, the test set image feature library is input into the obtained convolutional neural network model to verify the accuracy rate; when the accuracy rate reaches the standard, a second classification model is established. Otherwise, adjust the parameters and retrain until the accuracy rate reaches the standard.

[0133] It should be noted that the output layers of the first classification model and the second classification model both use soft labels, that is, they do not directly output the classification result of a certain waveform segment, but output the probability that the waveform segment is a certain type of event. At the same time, the classification type with the highest probability is used as the classification result of the convolutional neural network model for this waveform segment to verify the accuracy rate of the convolutional neural network model.

[0134] In addition, in this embodiment, a feedback mechanism based on the partition accuracy rate is proposed.

[0135] On the one hand, the causes of microseismic monitoring data are complex. Using existing manual recognition methods for pre-classification cannot guarantee 100% correct classification. The correct classification rate of manual recognition for typical waveforms is relatively high. However, for non-typical complex waveforms, due to the current research progress and the capabilities and experience of the personnel engaged in classification, there is still room for improvement in the correct classification rate of manual recognition, which may have a certain impact on the learning and training of the convolutional neural network model. On the other hand, due to the black-box nature of the convolutional neural network, its interpretability has always been a difficult point in the application process. At the same time, further research on non-typical complex waveforms not only involves the classification results of waveforms, but also requires understanding and interpreting the classification process. Therefore, it is considered to add feedback based on the analysis of the classification results in the convolutional neural network to interpret the classification process of the convolutional neural network and provide assistance for further analysis of waveforms.

[0136] First, assume that there are n (1, 2, 3, …, i, …, n) types of microseismic events. For a segment of waveform, the classification result after passing through the convolutional neural network is a soft label, which is a 1×N numerical matrix, where a 1i (i = 1, 2, 3, …, n) is the probability that this segment of waveform is classified as a microseismic event type i and . Let a 1i (i = 1, 2, 3, …, n ) be arranged in descending order. The maximum value is defined as Result1, and the subsequent values are defined as Result2, Result3, …, Resultn respectively; calculate the variance between Result1 and Result2, denoted as the two-item variance (i.e., the first variance); calculate the variance of Result1, Result2, and Result3, denoted as the three-item variance; calculate the variance of Result1, Result2, Result3, ……, Resulti, denoted as the i-item variance; calculate the variance of Result1, Result3, ……, Resultn, denoted as the overall variance (i.e., the second variance).

[0137] See Figure 8 as shown below:

[0138] ① When the two-item variance ≥ the overall variance, it is considered that this segment of waveform belongs to a typical waveform;

[0139] ② When the two-item variance < the overall variance, it is considered that this segment of waveform belongs to an atypical waveform;

[0140] ③ Further, when the i -1 item variance < the overall variance and the i item variance ≥ the overall variance, it is considered that this segment of waveform is i-1 type of hybrid waveform.

[0141] After the constructed convolutional neural network is applied to classify the current data set, the data set can be divided into two categories: typical waveforms and atypical waveforms.

[0142] See Figure 9 As shown, calculate the accuracy rates for the recognition situations within the ranges of typical waveforms and atypical waveforms respectively. The results mainly have the following three situations:

[0143] ① The accuracy rate for recognizing typical waveforms is lower than 95%, and the accuracy rate for recognizing atypical waveforms is lower than 95%: It indicates that the selected features are not typical, the weight distribution of the features in classification is improper, or there are problems with the class labels and pre-classification results. First, consider adjusting the weights of each feature in classification. If the accuracy rate of the re-adjusted model does not increase, it is necessary to check the settings of the class labels and the correctness of the pre-classification results.

[0144] ② The accuracy rate for classifying typical waveforms is higher than 95%, and the accuracy rate for recognizing atypical waveforms is lower than 95%: Consider that the pre-classification of atypical waveforms may be incorrect. Use the current recognition results as class labels for a control experiment. If the accuracy rate increases significantly, accept the current recognition results; if the accuracy rate decreases, maintain the pre-classification and adjust the parameters to re-train until the accuracy rates of both typical waveforms and atypical waveforms are relatively high.

[0145] ③ The accuracy rate for recognizing typical waveforms is higher than 95%, and the accuracy rate for recognizing atypical waveforms is higher than 95%: It indicates that the learning effect of this convolutional neural network model is good, and it is determined that the accuracy rate of this model meets the standard.

[0146] Step 7: Synthesize the output results of the dual convolutional neural network model, and calculate the comprehensive recognition result by using a weighted method (the weights here can be adjusted according to the classification effects of the two convolutional neural network models. For example, the weights are each 0.5). Analyze and evaluate the comprehensive recognition result of the test set. When the accuracy rate meets the standard, save the dual convolutional neural network model; otherwise, return to Step 6 to adjust the parameters and re-establish the dual convolutional neural network model.

[0147] This embodiment does not require pre-filtering the initial microseismic monitoring waveforms. The process is simple and highly accurate. With this method, it is possible to automatically classify waveform data such as blasting events, instability and creep, rock-breaking construction disturbances, rock mass fracture microseismic events, invalid noise, or undefined waveforms in microseismic monitoring, providing important technical support for the efficient interpretation of rock mass damage microseismic data and stability evaluation.

[0148] The following illustrates this solution with actual examples.

[0149] (1) Obtain five different waveforms A, B, C, D, and E, as Figure 10 shown.

[0150] (2) Obtain the time-domain features of each waveform, including: duration, maximum amplitude value, minimum amplitude value, and average amplitude value, as shown in Table 1.

[0151] Table 1 Time-domain Feature Table

[0152]

[0153] (3) Extract the time-frequency feature images of each waveform, including: fast Fourier transform spectrum diagram, short-time Fourier transform time-frequency diagram plotted with an adaptive color scale, short-time Fourier transform time-frequency diagram plotted with a unified scale color scale, wavelet transform time-frequency diagram plotted with an adaptive color scale, and wavelet transform time-frequency diagram plotted with a unified scale color scale.

[0154] (4) Plot the amplitude distribution curves of waveforms A, B, C, D, and E, as Figure 11 shown. Calculate the statistical features of the amplitude distribution according to the amplitude distribution curve, as shown in Table 2.

[0155] Table 2 Amplitude Distribution Statistical Feature Table

[0156]

[0157] (5) Combine the time-domain features (duration, maximum amplitude value, minimum amplitude value, average amplitude value) of each waveform segment, time-frequency feature images (fast Fourier transform spectrum diagram, short-time Fourier transform time-frequency diagram, wavelet transform time-frequency diagram, etc.), and statistical features based on the amplitude distribution curve (maximum value, variance, standard deviation, root mean square, root mean amplitude, kurtosis, skewness, waveform factor, peak factor, impulse factor, margin factor, etc.) to obtain a waveform feature library.

[0158] (6) Based on the actual situation and existing classification methods, pre-classify each waveform involved in the waveform feature database and add category labels to form a dataset for constructing a dual convolutional neural network model. The sample size of each type of event should be equal. Suppose 2500 waveforms are obtained, and the event category labels are 1 - blasting event, 2 - instability and creep, 3 - rock-breaking construction disturbance, 4 - microseismic event of rock mass fracture, 5 - invalid noise or undefined waveform, a total of five types of event types, and the number of each type of event is equal.

[0159] Taking the above waveforms A, B, C, D, and E as an example, their category labels are shown in Table 3.

[0160] Table 3 Category Label Table

[0161]

[0162] (7) Divide the dataset. Divide the dataset into a training set and a test set, and further divide them into a quantization database and an image database. That is, form: a training set quantization database, a training set image database, a test set quantization database, and a test set image database.

[0163] Specifically (taking the total number of waveforms as 2500 segments as an example):

[0164] Mix and shuffle the dataset used to construct the dual convolutional neural network model, and extract 70% (1750 waveform segments) of it as the training set and 30% (750 waveform segments) as the test set. For each waveform segment, summarize and tile the quantization parameter features of the waveform (duration, maximum amplitude value, minimum amplitude value, average amplitude value, and maximum value, variance, standard deviation, root mean square, root mean square amplitude, kurtosis, skewness, waveform factor, peak factor, impulse factor, margin factor, etc. based on the amplitude distribution curve) into a one-dimensional array; for the 1750 waveform segments in the training set, generate an excel table, with each row representing a waveform segment, where the first column is the serial number (or name) of the waveform segment, the second column is the class label of the waveform segment, and the 3rd to 17th columns are the quantization parameter features of the waveform segment. The excel is named the training set quantization database; for the 750 waveform segments in the test set, generate an excel table, with each row representing a waveform segment, where the first column is the serial number (or name) of the waveform segment, the second column is the class label of the waveform segment, and the 3rd to 17th columns are the quantization parameter features of the waveform segment. The excel is named the test set quantization database.

[0165] For each waveform segment, use MATLAB to implement the tiling of time-frequency feature images, forming a feature image with the original waveform diagram in the upper left, the fast Fourier transform frequency spectrum diagram in the middle left, the short-time Fourier transform time-frequency diagram (adaptive color scale) in the upper right, the short-time Fourier transform time-frequency diagram (uniform scale color scale) in the middle right, the wavelet transform time-frequency diagram (adaptive color scale) in the lower left, and the wavelet transform time-frequency diagram (uniform scale color scale) in the lower right; after all processing is completed, 2500 feature images of equal length, equal width, and equal resolution are formed, ensuring that the name of each feature image is unique; establish a training set image database folder, and create sub-folders under it according to the number of classified categories, and classify and place the 1750 feature images in the training set according to the class label; establish a test set image database folder, and create sub-folders under it according to the number of classified categories, and classify and place the 750 feature images in the test set according to the class label. Taking the aforementioned waveform A as an example, the feature image is as Figure 12 shown, and the image is 2188 pixels high and 1133 pixels wide.

[0166] (8) Construct a dual convolutional neural network model to learn and train the waveform features, and form and output a dual convolutional neural network model for classification.

[0167] (9) Apply the established dual convolutional neural network model to classify unknown waveforms.

[0168] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The order of execution of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0169] Figure 13 It is a schematic structural diagram of the rock mass microseismic identification device 130 provided by the embodiment of the present invention. Refer to Figure 13 As shown, the device includes:

[0170] An acquisition module 131, configured to acquire the original microseismic monitoring waveform and segment it to obtain multiple segments of waveforms.

[0171] A feature extraction module 132, configured to extract time-domain features, amplitude distribution statistical features, and time-frequency feature images for each segment of waveform.

[0172] An identification module 133, configured to input the time-domain features and amplitude distribution statistical features into the first classification model in the pre-trained dual convolutional neural network model to obtain a first classification result; splice the time-frequency feature images and input them into the second classification model in the dual convolutional neural network model to obtain a second classification result; and determine the microseismic event type based on the first classification result and the second classification result.

[0173] As a possible implementation manner, the time-domain features include at least one of the following:

[0174] Event effective duration, maximum amplitude value, minimum amplitude value, and average amplitude value;

[0175] Among them, the event effective duration is calculated by the following method:

[0176] Calculate the short-term and long-term average ratio data of this segment of waveform;

[0177] According to determine the initial arrival threshold, and according to determine the termination threshold; where a 、 b are preset coefficients and are both less than 1; is the difference between the maximum value and the minimum value of the short-term and long-term average ratio; Z is the mode in the short-term and long-term average ratio;

[0178] Determine the arrival time as the moment when the ratio of the long-term average value to the short-term average value is greater than the arrival threshold for the first time, and determine the termination time as the moment when the ratio of the long-term average value to the short-term average value is less than the termination threshold for the first time since the arrival time;

[0179] Based on the arrival time and the termination time, determine the effective duration of the event.

[0180] As a possible implementation, the time-frequency feature image includes at least one of the following:

[0181] Fast Fourier transform spectrum diagram, short-time Fourier transform time-frequency diagram plotted with an adaptive color scale, short-time Fourier transform time-frequency diagram plotted with a unified scale color scale, wavelet transform time-frequency diagram plotted with an adaptive color scale, and wavelet transform time-frequency diagram plotted with a unified scale color scale.

[0182] As a possible implementation, extract the amplitude distribution statistical features, including:

[0183] Calculate the number of sampling points corresponding to different amplitudes in this section of the waveform, fit to obtain the amplitude distribution curve, and calculate the amplitude distribution statistical features according to the amplitude distribution curve;

[0184] Among them, the amplitude distribution statistical features include at least one of the following:

[0185] Maximum value, variance, standard deviation, root mean square, root mean square amplitude, kurtosis, skewness, waveform factor, peak factor, impulse factor, margin factor.

[0186] As a possible implementation, both the first classification result and the second classification result are the probabilities of various microseismic event types;

[0187] Based on the first classification result and the second classification result, determine the microseismic event type, including:

[0188] For each microseismic event type, perform weighted summation on the probability in the first classification result and the probability in the second classification result to obtain the comprehensive probability of this microseismic event type;

[0189] Determine the microseismic event type with the maximum comprehensive probability as the microseismic event type of this section of the waveform.

[0190] As a possible implementation, the dual convolutional neural network model is trained in the following way:

[0191] Obtain different sections of waveforms and divide them into a training set and a test set;

[0192] Train and test the first convolutional neural network model through the time domain features, amplitude distribution statistical features of different sections of waveforms in the training set and the test set, and the predetermined microseismic event types to obtain the first classification model;

[0193] Train and test the second convolutional neural network model using the time-frequency feature images of different segments of waveforms in the training set and the test set to obtain the second classification model;

[0194] Determine the weights of the first classification model and the second classification model to obtain the double convolutional neural network model to be verified;

[0195] Judge whether the recognition accuracy rate of the double convolutional neural network model to be verified for the test set reaches the preset standard. If it reaches, determine it as the double convolutional neural network model.

[0196] As a possible implementation, training and testing the first convolutional neural network model or the second convolutional neural network model includes:

[0197] After training the target convolutional neural network model, determine the waveform type of each segment of waveform according to the classification result of each segment of waveform in the test set output by the target convolutional neural network model; wherein, the target convolutional neural network model is the first convolutional neural network model or the second convolutional neural network model, and the waveform types include typical waveforms and atypical waveforms;

[0198] Calculate the recognition accuracy rates of the target convolutional neural network model for typical waveforms and atypical waveforms respectively. If the recognition accuracy rates of the target convolutional neural network model for typical waveforms and atypical waveforms are both greater than the preset accuracy rate threshold, determine that the training of the target convolutional neural network model is completed;

[0199] If the classification accuracy rate of the target convolutional neural network model for typical waveforms and / or atypical waveforms is less than or equal to the accuracy rate threshold, adjust the parameters, training set or test set of the target convolutional neural network model, and retrain or test the target convolutional neural network model.

[0200] As a possible implementation, the classification result is the probability of various microseismic event types;

[0201] Determine the waveform type of each segment of waveform according to the classification result of each segment of waveform in the test set output by the target convolutional neural network model, including:

[0202] Sort various microseismic event types in descending order of probability;

[0203] Calculate the variance of the probabilities of the first two microseismic event types in the sequence to obtain the first variance;

[0204] Calculate the variance of the probabilities of all microseismic event types in the sequence to obtain the second variance;

[0205] If the first variance is greater than or equal to the second variance, determine that this segment of waveform is a typical waveform;

[0206] If the first variance is less than the second variance, it is determined that this section of waveform is an atypical waveform.

[0207] Figure 14 It is a schematic diagram of the electronic device 140 provided by an embodiment of the present invention. As Figure 14 shown, the electronic device 140 of this embodiment includes: a processor 141, a memory 142, and a computer program 143 stored in the memory 142 and executable on the processor 141, such as a rock mass microseismic identification program. When the processor 141 executes the computer program 143, the steps in the above-mentioned various embodiments of the rock mass microseismic identification method are implemented, such as Figure 1 the steps S101 to S105 shown. Alternatively, when the processor 141 executes the computer program 143, the functions of each module in the above-mentioned device embodiments are implemented, such as Figure 13 the functions of the modules 131 to 133 shown.

[0208] Exemplarily, the computer program 143 can be divided into one or more modules / units. One or more modules / units are stored in the memory 142 and executed by the processor 141 to complete the present invention. One or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program 143 in the electronic device 140.

[0209] The electronic device 140 can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The electronic device 140 may include, but is not limited to, a processor 141 and a memory 142. Those skilled in the art can understand that Figure 14 this is only an example of the electronic device 140, and does not constitute a limitation on the electronic device 140. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the electronic device 140 may further include input / output devices, network access devices, a bus, etc.

[0210] The so-called processor 141 may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor, etc.

[0211] The memory 142 can be an internal storage unit of the electronic device 140, such as the hard disk or memory of the electronic device 140. The memory 142 can also be an external storage device of the electronic device 140, such as a plug-in hard disk equipped on the electronic device 140, a Smart Media Card (SMC), a Secure Digital (SD) card, a FlashCard, etc. Further, the memory 142 can also include both the internal storage unit of the electronic device 140 and the external storage device. The memory 142 is used to store computer programs and other programs and data required by the electronic device 140. The memory 142 can also be used to temporarily store the data that has been output or will be output.

[0212] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the above-mentioned division of each functional unit and module is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiments can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of the present application. The specific working processes of the units and modules in the above system can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0213] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0214] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0215] In the embodiments provided by the present invention, it should be understood that the disclosed device / electronic device and method can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical or other forms.

[0216] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0217] In addition, in each embodiment of the present invention, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0218] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above method embodiments of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0219] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for identifying microseisms in rock masses, characterized in that, Including: Obtain the original microseismic monitoring waveform and segment it to obtain multiple segments of waveforms; For each segment of waveform, extract time-domain features, amplitude distribution statistical features, and time-frequency feature images; The time-frequency feature images include at least one of the following: fast Fourier transform spectrogram, short-time Fourier transform time-frequency diagram plotted with an adaptive color scale, short-time Fourier transform time-frequency diagram plotted with a unified scale color scale, wavelet transform time-frequency diagram plotted with an adaptive color scale, and wavelet transform time-frequency diagram plotted with a unified scale color scale; Input the time-domain features and the amplitude distribution statistical features into the first classification model in a pre-trained dual convolutional neural network model to obtain a first classification result; Input the stitched time-frequency feature images into the second classification model in the dual convolutional neural network model to obtain a second classification result; tile each time-frequency feature image to form a feature image with the original microseismic monitoring waveform in the upper left, fast Fourier transform spectrogram in the upper middle, short-time Fourier transform time-frequency diagram in the upper right, short-time Fourier transform time-frequency diagram in the middle right, wavelet transform time-frequency diagram in the lower left, and wavelet transform time-frequency diagram in the lower right; Based on the first classification result and the second classification result, determine the microseismic event type.

2. The microseismic identification method for rock mass according to claim 1, characterized in that The time-domain features include at least one of the following: Event effective duration, maximum amplitude value, minimum amplitude value, and average amplitude value; Among them, the event effective duration is calculated by the following method: Calculate the long-short time average ratio data of this segment of waveform; According to λ a = a×Δd + Z to determine the first arrival threshold, and according to λ b = b×Δd + Z to determine the termination threshold; where a and b are preset coefficients and both are less than 1; Δd is the difference between the maximum value and the minimum value of the long-term and short-term average ratio; Z is the mode in the long-term and short-term average ratio; Determine the arrival time when the long-short time average ratio is first greater than the arrival threshold, and determine the termination time when the long-short time average ratio is first less than the termination threshold since the arrival time; Based on the arrival time and the termination time, determine the event effective duration.

3. The microseismic identification method of rock mass according to claim 1, characterized in that, Extract amplitude distribution statistical features, including: Calculate the number of sampling points corresponding to different amplitudes in this segment of waveform, fit to obtain an amplitude distribution curve, and calculate amplitude distribution statistical features according to the amplitude distribution curve; Among them, the amplitude distribution statistical features include at least one of the following: Maximum value, variance, standard deviation, root mean square, root mean square amplitude, kurtosis, skewness, waveform factor, peak factor, impulse factor, margin factor.

4. The microseismic identification method of rock mass according to claim 1, characterized in that Both the first classification result and the second classification result are probabilities of various microseismic event types; The determining the microseismic event type based on the first classification result and the second classification result includes: For each microseismic event type, perform weighted summation on the probability in the first classification result and the probability in the second classification result to obtain the comprehensive probability of this microseismic event type; Determine the microseismic event type with the maximum comprehensive probability as the microseismic event type of this segment of waveform.

5. The method for identifying microseisms in rock masses according to any one of claims 1 to 4, characterized in that, The dual convolutional neural network model is trained by the following method: Obtain different segments of waveforms and divide them into a training set and a test set; Train and test the first convolutional neural network model through the time-domain features, amplitude distribution statistical features of different segments of waveforms in the training set and the test set, and the pre-determined microseismic event types to obtain the first classification model; Train and test the second convolutional neural network model using the time-frequency feature images of different segments of waveforms in the training set and the test set to obtain the second classification model; Determine the weights of the first classification model and the second classification model to obtain the dual convolutional neural network model to be verified; Judge whether the recognition accuracy of the dual convolutional neural network model to be verified for the test set reaches the preset standard. If it reaches, determine it as the dual convolutional neural network model.

6. The method for identifying microseisms in rock masses according to claim 5, wherein, Training and testing the first convolutional neural network model or the second convolutional neural network model includes: After training the target convolutional neural network model, determine the waveform type of each segment of waveform in the test set according to the classification result of each segment of waveform output by the target convolutional neural network model; wherein, the target convolutional neural network model is the first convolutional neural network model or the second convolutional neural network model, and the waveform types include typical waveforms and atypical waveforms; Calculate the recognition accuracy of the target convolutional neural network model for typical waveforms and atypical waveforms respectively. If the recognition accuracy of the target convolutional neural network model for both typical waveforms and atypical waveforms is greater than the preset accuracy threshold, determine that the target convolutional neural network model training is completed; If the classification accuracy of the target convolutional neural network model for typical waveforms and / or atypical waveforms is less than or equal to the accuracy threshold, adjust the parameters of the target convolutional neural network model, the training set or the test set, and retrain or test the target convolutional neural network model.

7. The microseismic identification method for rock mass according to claim 6, characterized in that, The classification result is the probability of various microseismic event types; The determining the waveform type of each segment of waveform according to the classification result of each segment of waveform in the test set output by the target convolutional neural network model includes: Sort various microseismic event types in descending order of probability; Calculate the variance of the probabilities of the first two microseismic event types in the sequence to obtain the first variance; Calculate the variance of the probabilities of all microseismic event types in the sequence to obtain the second variance; If the first variance is greater than or equal to the second variance, determine that this segment of waveform is a typical waveform; If the first variance is less than the second variance, determine that this segment of waveform is an atypical waveform.

8. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images