A microseismic event hazard classification method considering the spatiotemporal characteristics of monitoring data

Abstract

The present invention relates to a microseismic event hazard classification method considering the spatiotemporal characteristics of monitoring data. Different characteristic indicators are established by considering the spatiotemporal characteristics of microseismic monitoring data, and then data imbalance and One-Hot coding are processed. The processed indicators are used as features, and whether dangerous disasters such as rock burst or surface vibration occur as labels. Then, a hazard classification model of microseismic events is established through a one-dimensional convolutional neural network. The classification method makes full use of the spatiotemporal characteristics of monitoring data, has fast training time, and good classification effect, and solves the problem that most current studies rely too much on the type of monitoring data and ignore the spatiotemporal characteristics of microseismic monitoring data in the classification of microseismic event hazards in mines.

Description

Technical Field

[0001] The present invention relates to the field of dangerous disasters such as rock burst in mines, and in particular to a method for classifying the hazard of microseismic events taking into account the temporal and spatial characteristics of monitoring data. Background Art

[0002] The occurrence of dangerous disasters such as rock burst in mines often causes serious damage to tunnels and heavy casualties of workers, and may even cause ground vibrations and small earthquakes. Preventing and controlling the occurrence of dangerous events in mines requires accurate and reliable analysis of the danger of microseismic events, and classifying microseismic events according to the degree of danger, so that appropriate measures can be taken to eliminate the threat. In the prior art, many scholars have analyzed and predicted the danger of microseismic events based on microseismic monitoring data, but in the process of establishing analysis indicators, they rely on monitoring values ​​such as energy and electromagnetic radiation signals, without considering the spatiotemporal characteristics of the data, and have high requirements for microseismic monitoring data (many indicators are considered and the monitoring of indicators is difficult), but they cannot fully utilize the relationship between data, which increases the monitoring cost, and some methods are complex in design and difficult to be directly applied to engineering. For example, the microseismic monitoring data in most mines mainly includes time series, location and energy data, and there is no stress monitoring data or the stress monitoring data is difficult to accurately map to a certain point, but stress is used as an analysis indicator in most studies. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention proposes a microseismic event hazard classification method considering the spatiotemporal characteristics of monitoring data, comprising the following steps:

[0004] S1. Establish characteristic indicators for classification of microseismic event hazards

[0005] The energy of this microseismic event, the time density of n consecutive microseismic events before this microseismic event, and the spatial density of n consecutive microseismic events before this microseismic event are used as characteristic indicators for the classification of microseismic event hazards; the time density is represented by the mean of the time intervals between n consecutive microseismic events, and the spatial density is represented by the variance of the distances between the locations where n consecutive microseismic events occurred and the center of the location where these n microseismic events occurred;

[0006] S2, establish a data set and handle sample imbalance

[0007] In M consecutive microseismic events, the monitoring data of the mth to m+nth microseismic events are used as the data source of a sample in the data set, 1≤m≤Mn, and at most Mn samples are established, each sample includes feature data and label data, the feature data includes the time density of the mth to m+n-1th microseismic events, the spatial density of the mth to m+n-1th microseismic events, and the energy value of the m+nth microseismic event; the label data is the danger category of the m+nth microseismic event, which is divided into dangerous events and non-dangerous events, and the label data is processed by One-Hot encoding;

[0008] Sample imbalance processing: During the sampling process, the feature data of samples with dangerous event labels in the data set are slightly disturbed and then replicated to balance the number of samples with different labels.

[0009] Preferably, after One-Hot encoding processing, the label corresponding to the dangerous event is 1, and the label corresponding to the non-dangerous event is 0.

[0010] Preferably, the method of slight perturbation is to add random numbers.

[0011] S3, establish a microseismic event hazard classification model

[0012] A three-layer classification model is established based on a one-dimensional convolutional neural network. Each layer consists of a convolutional layer, a batch normalization layer, and an activation layer. The Relu function is used as the activation function. At the end of the classification model, two fully connected layers are set for classification, and finally the Softmax function is used to output the classification label.

[0013] Preferably, the function of the convolution layer is information fusion and extraction, the number of convolution kernels of the three convolution layers are 6, 12 and 12 respectively, the size of the convolution kernel is 1×1, and the number of neurons in the two fully connected layers is 4 and 2 respectively.

[0014] Preferably, during the training process, the Adam optimizer is selected to back-propagate gradient information to update the parameters of the network, and the MSELoss loss function is used to represent the degree of difference between the model calculation results and the actual data.

[0015] S4. Evaluating the performance of microseismic event hazard classification models

[0016] The performance of the microseismic event hazard classification model is evaluated by using a confusion matrix or establishing two test sets containing mixed label samples and separate hazard event label samples;

[0017] S5, Application of microseismic event hazard classification

[0018] The energy of this microseismic event, the time density of n consecutive microseismic events before this microseismic event, and the spatial density of n consecutive microseismic events before this microseismic event are used as input data, substituted into the microseismic event hazard classification model, and the label data is output to determine whether the category of this microseismic event is a dangerous event or a non-dangerous event.

[0019] Beneficial effects: The present invention relates to a microseismic event hazard analysis (classification) method that considers the spatiotemporal characteristics of monitoring data. Different characteristic indicators are established by considering the spatiotemporal characteristics of microseismic monitoring data, and then data imbalance and One-Hot encoding processing are performed. The processed indicators are used as features, and whether a rock burst event or a strong vibration occurs is used as a label. Then, a classification model for microseismic events is established through a one-dimensional convolutional neural network. This classification method makes full use of the spatiotemporal characteristics of monitoring data, has a fast training time, and has a good classification effect. It solves the problem that most current studies rely too much on the type of monitoring data and ignore the spatiotemporal characteristics of microseismic monitoring data in the risk analysis of microseismic events. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] FIG. 1( a ) is a flow chart of step S1 of the method for classifying the risk of microseismic events according to the present invention;

[0021] FIG1( b ) is a flow chart of step S2 of the method for classifying the risk of microseismic events according to the present invention;

[0022] FIG1( c ) is a flow chart of step S3 of the method for classifying the risk of microseismic events according to the present invention;

[0023] Figure 2 Microseismic event hazard classification model based on convolutional neural network (B is the batch size);

[0024] Figure 3 An example of evaluating the performance of a microseismic event hazard classification model using microseismic monitoring data from a mine. DETAILED DESCRIPTION

[0025] The technical solution of the present invention is described in more detail below in conjunction with the accompanying drawings.

[0026] As shown in Figure 1(a), Figure 1(b), Figure 1(c), Figure 2 As shown, a microseismic event hazard classification method considering the spatiotemporal characteristics of monitoring data is used to determine whether a microseismic event that has occurred is a dangerous event or a non-dangerous event, including the following steps:

[0027] S1. Establish characteristic indicators for classification of microseismic event hazards

[0028] After studying the microseismic monitoring data of different mines, it was found that before the occurrence of dangerous events such as rock burst, the surrounding rock system was in an unstable adjustment period due to the energy exchange between the surrounding rock system and the outside world, so there was an obvious microseismic active period. This microseismic active period was mainly manifested in the more concentrated time and space density of multiple consecutive microseismic events before the occurrence of dangerous disasters such as rock burst. For example, the time intervals of these multiple consecutive microseismic events were small and the spatial positions were relatively concentrated. And through research, it was found that the energy of this microseismic event is also an important indicator for classifying dangerous events such as rock burst.

[0029] Based on this, the energy of this microseismic event, the time density of the n consecutive microseismic events before this microseismic event, and the spatial density of the n consecutive microseismic events before this microseismic event are determined as characteristic indicators for the classification of microseismic event hazards; the time density is represented by the mean of the time intervals (time differences) between the n consecutive microseismic events, and the spatial density is represented by the variance of the distances between the locations where the n consecutive microseismic events occurred and the center of the location where these n microseismic events occurred;

[0030] S2, establish a data set and handle sample imbalance

[0031] In M consecutive microseismic events, the monitoring data of the mth to m+nth microseismic events are used as the data source of a sample in the data set, 1≤m≤Mn, then at most Mn samples can be established, each sample includes characteristic data and label data, wherein the characteristic data includes the time density of the mth to m+n-1th microseismic events (the mean of the time intervals between each other), the spatial density of the mth to m+n-1th microseismic events (the variance of the distances between the respective occurrence locations of the n consecutive microseismic events and the center of the occurrence location of these n microseismic events) and the energy value of the m+nth microseismic event; the label data is the danger category of the m+nth microseismic event, that is, the label data is whether a dangerous event (impact ground pressure or ground vibration) has occurred, and the label data needs to be processed by One-Hot encoding, and after processing, the label corresponding to the dangerous event is 1, and the label corresponding to the non-dangerous event is 0;

[0032] For example, in 7 consecutive microseismic events, the monitoring data of the mth to m+5th microseismic events are used as the data source of a sample in the data set, 1≤m≤7-5, then at most 2 data can be established; the characteristic data of the first sample include the time density of the 1st to 5th microseismic events, the spatial density of the 1st to 5th microseismic events and the energy value of the 6th microseismic event, and the label data is the hazard category of the 6th microseismic event; the characteristic data of the second sample include the time density of the 2nd to 6th microseismic events, the spatial density of the 2nd to 6th microseismic events and the energy value of the 7th microseismic event, and the label data is the hazard category of the 7th microseismic event.

[0033] Sample imbalance processing: The number of dangerous events in mine microseismic events is much smaller than the number of non-dangerous events. Therefore, the upsampling method is used. During the sampling process, the feature data of the sample labels of dangerous events in the data set are slightly disturbed (such as adding random numbers) and then replicated to balance the number of samples with different labels (dangerous categories);

[0034] S3, establish a microseismic event hazard classification model

[0035] A three-layer classification model is established based on a one-dimensional convolutional neural network. Each layer consists of a convolutional layer, a batch normalization layer, and an activation layer. The function of the convolutional layer is to fuse and extract information. The number of convolution kernels in the three convolutional layers is 6, 12, and 12 respectively, and the size of the convolution kernel is 1×1. The Relu function is used as the activation function. At the end of the classification model, two fully connected layers are set for classification. The number of neurons is 4 and 2 respectively. Finally, the Softmax function is used to output the classification label.

[0036] Preferably, during the training process, the Adam optimizer is selected to back-propagate gradient information to update the parameters of the network, and the MSELoss loss function is used to represent the degree of difference between the model calculation results and the actual data.

[0037] S4. Evaluating the performance of microseismic event hazard classification models

[0038] Considering the importance of correctly classifying dangerous events such as rock burst, the performance of the microseismic hazard classification model is evaluated by using a confusion matrix or establishing two test sets containing mixed label samples and separate dangerous event label samples;

[0039] like Figure 3 As shown in the figure, the performance of the classification model was evaluated based on the microseismic event monitoring data of a mine. Training samples were randomly selected and 20 experiments were carried out. In each experiment, the training model achieved the highest accuracy on the test set containing mixed label samples and the test set containing dangerous event label samples separately. The highest accuracy rates were 0.957 and 0.893 respectively.

[0040] S5, Application of microseismic event hazard classification

[0041] The energy of this microseismic event, the time density of n consecutive microseismic events before this microseismic event, and the spatial density of n consecutive microseismic events before this microseismic event are used as input data and substituted into the microseismic event hazard classification model to output label data, that is, the category of this microseismic event is a dangerous event or a non-dangerous event.

Claims

1. A microseismic event hazard classification method considering the spatiotemporal characteristics of monitoring data, characterized in that: The steps include: S1. Establish characteristic indicators for classification of microseismic event hazards Based on the energy of this microseismic event and the continuous n The time density of microseismic events and the continuous n The spatial density of microseismic events is used as a characteristic index for the classification of microseismic event hazards; The time intensity is continuous n The spatial density is expressed as the mean of the time intervals between each microseismic event. n The locations of the microseismic events are n The variance of the distance between the centers of the microseismic events is expressed; S2, establish a data set and handle sample imbalance In continuous M Among the microseismic monitoring events, m From m + n The monitoring data of microseismic events is used as the data source of a sample in the dataset, 1≤ m ≤ M - n , at most M - n samples, each sample includes feature data and label data, the feature data includes m From m + n -1 The time density of microseismic events, m From m + n -1 The spatial density of microseismic events and m + n Energy value of microseismic events; The label data is m + n The danger categories of microseismic events are divided into dangerous events and non-dangerous events, and the label data is processed by One-Hot encoding; Sample imbalance processing: During the sampling process, the feature data of samples with dangerous event labels in the data set are slightly disturbed and then replicated to balance the number of samples with different labels. S3, establish a microseismic event hazard classification model A three-layer classification model is established based on a one-dimensional convolutional neural network. Each layer consists of a convolutional layer, a batch normalization layer, and an activation layer. The Relu function is used as the activation function. At the end of the classification model, two fully connected layers are set for classification, and finally the Softmax function is used to output the classification label. S4. Evaluating the performance of microseismic event hazard classification models The performance of the microseismic event hazard classification model is evaluated by using a confusion matrix or establishing two test sets containing mixed label samples and separate hazard event label samples; S5, Application of microseismic event hazard classification Based on the energy of this microseismic event and the continuous n The time density of microseismic events and the continuous n The spatial density of microseismic events is used as input data, substituted into the microseismic event hazard classification model, and the label data is output to determine whether the category of this microseismic event is a dangerous event or a non-dangerous event.

2. The microseismic event hazard classification method according to claim 1, characterized in that: In step S2, after One-Hot encoding processing, the label corresponding to the dangerous event is 1, and the label corresponding to the non-dangerous event is 0.

3. The microseismic event hazard classification method according to claim 1 or 2, characterized in that: In step S2, the slight disturbance method is to add a random number.

4. The microseismic event risk classification method according to claim 1, characterized in that: In step S3, the function of the convolution layer is to fuse and extract information. The number of convolution kernels of the three convolution layers are 6, 12 and 12 respectively, the size of the convolution kernel is 1×1, and the number of neurons in the two fully connected layers is 4 and 2 respectively.

5. The microseismic event hazard classification method according to claim 1 or 4, characterized in that: In step S3, the Adam optimizer is selected during the training process to back-propagate gradient information to update the network parameters, and the MSELoss loss function is used to represent the degree of difference between the model calculation results and the actual data.

Images