An intelligent discrimination method for rock failure modes based on unique microseismic signals

Abstract

The present invention provides an intelligent method for distinguishing rock failure modes based on unique microseismic signals, which includes collecting the original microseismic waveforms of known rock failure modes, extracting the characteristic parameters in the time domain and frequency domain of the microseismic waveforms by using Fourier transform, performing deviation standardization on the extracted characteristic parameters and then flattening the characteristic parameters into a one-dimensional microseismic parameter feature map, constructing a rock failure mode discrimination network with a convolutional neural network as the basic framework, training and learning the mapping relationship between the one-dimensional microseismic parameter feature map and the rock failure mode discrimination to construct an intelligent rock failure mode discrimination model, verifying the intelligent rock failure mode discrimination model, and inputting the processed microseismic waveforms of unknown failure modes into the trained intelligent rock failure mode discrimination model for failure mode discrimination. This application can complete the discrimination of rock failure modes only by using a single microseismic signal, without using the variation law of microseismic signals within a continuous time, reducing the subjective and objective errors in the monitoring and early warning of dynamic disasters.

Description

Technical Field

[0001] The present invention relates to the technical field of mine microseismic monitoring and early warning, and particularly relates to an intelligent method for distinguishing rock failure modes based on unique microseismic signals. Background Art

[0002] Microseismic monitoring is commonly used to monitor dynamic disasters such as rock bursts and coal and gas outbursts. By using seismographs to monitor the elastic waves generated during the stress deformation or fracture process of rock masses, the "time-space-intensity" information of microseismic events is obtained to predict the occurrence of dynamic disasters such as rock bursts. Judging the intensity of dynamic disasters based on the energy of microseismic events is one of the core issues in microseismic monitoring.

[0003] Currently, conventional microseismic monitoring and early warning methods generally analyze the signals monitored by the microseismic monitoring system uniformly and judge the risk level of dynamic disasters according to the variation law of continuous time. However, the inventors of this application have found through research that this method has two defects: First, traditional early warning technologies received by the microseismic monitoring system analyze these signals uniformly to judge their risk levels. However, a mine has a complex geological environment, and rocks can also have different failure modes. The waveform evolution laws of microseismic signals of rock failures in different failure modes are also different. Therefore, analyzing all signals uniformly will cause subjective errors in the monitoring and early warning of dynamic disasters. Second, in mines, rock failure signals may be unable to obtain data on the entire process of rock failure due to attenuation and sensor failures, etc., which will cause objective errors in monitoring and early warning.

[0004] In summary, under the complex geological environment conditions of mines, traditional microseismic monitoring and early warning methods have certain limitations in distinguishing failure modes and lithologies, etc. Currently, the microseismic monitoring and early warning methods for complex mine environment conditions are still affected by unclear monitoring signals and discontinuous monitoring signals. Summary of the Invention

[0005] Aiming at the technical problems in the existing microseismic monitoring and early warning methods that because rocks can have different failure modes, the waveform evolution laws of microseismic signals of rock failures in different failure modes are also different, analyzing all signals uniformly will cause subjective errors in the monitoring and early warning of dynamic disasters, and in mines, rock failure signals may be unable to obtain data on the entire process of rock failure due to attenuation and sensor failures, etc., which will cause objective errors in monitoring and early warning, the present invention provides an intelligent method for distinguishing rock failure modes based on unique microseismic signals.

[0006] To solve the above technical problems, the present invention adopts the following technical solutions:

[0007] An intelligent method for distinguishing rock failure modes based on unique microseismic signals, comprising the following steps:

[0008] S1. Collect the original microseismic waveforms of known rock failure modes;

[0009] S2. Use Fourier transform to extract the characteristic parameters in the time domain and frequency domain of the microseismic waveforms. The characteristic parameters include amplitude, duration, rise time, ring count, rise count, energy, effective voltage, average level, centroid frequency domain, and peak frequency;

[0010] S3. Perform deviation standardization on the characteristic parameters in the time domain and frequency domain of the extracted microseismic waveforms, and tile the standardized data of the characteristic parameters into a one-dimensional microseismic parameter feature map;

[0011] S4. Construct a rock failure mode discrimination network M_Net with a convolutional neural network as the basic framework;

[0012] S5. Use the one-dimensional microseismic parameter feature map as the input feature, the known rock failure modes as labels, and the rock failure mode discrimination network M_Net as the basic framework to train and learn the mapping relationship between the one-dimensional microseismic parameter feature map and the discrimination of rock failure modes, and construct an intelligent rock failure mode discrimination model;

[0013] S6. Verify the intelligent rock failure mode discrimination model: Based on the discrimination accuracy rate of the rock failure mode, when the accuracy rate is less than the preset value, adjust the hyperparameters of the intelligent rock failure mode discrimination model, and retrain the model according to step S5 until the accuracy rate is greater than or equal to the preset value, then it is considered that the model has been trained well and can be applied;

[0014] S7. Process the microseismic waveforms of unknown failure modes according to steps S2 and S3 above, and then input them into the trained intelligent rock failure mode discrimination model in step S5 for failure mode discrimination, realizing the application of discriminating microseismic waveforms of unknown rock failure modes.

[0015] Furthermore, the Fourier transform in step S2 adopts the following formula:

[0016]

[0017] where F(ω) represents the spectrum after Fourier transform, f(t) is the original signal, e -iωt is the complex exponential function, and ω is the frequency.

[0018] Furthermore, the deviation standardization in step S3 adopts the following formula:

[0019]

[0020] where x new is the result after standardization, x is the data to be standardized, x maxis the maximum value of the same type of eigenvalue data, x min is the maximum value of the same type of eigenvalue data.

[0021] Furthermore, the rock failure mode discrimination network M_Net constructed in step S4 includes four sequentially connected convolutional blocks. Each convolutional block includes a convolutional layer, a batch normalization layer, an activation layer, and a max pooling layer, which are connected in sequence from the input end to the output end. The convolutional kernel size of the convolutional layer in each convolutional block is 1×1. The number of convolutional kernels in the convolutional layers of the four convolutional blocks is 32, 64, 128, and 256 in sequence. The pooling kernel size of the max pooling layer in the four convolutional blocks is 1×1, and the pooling stride is 1.

[0022] Furthermore, the convolution calculation formula of the convolutional layer is as follows:

[0023]

[0024] where I(i - m, j - n) is the input signal, K(m, n) is the convolutional kernel, i, j are the indices of the output feature map, and m, n are the indices of the convolutional kernel.

[0025] Furthermore, the implementation of the batch normalization layer includes the following steps:

[0026] S41. Calculate the mean of the microseismic feature parameters of each training batch;

[0027] S42. Calculate the variance of the microseismic feature parameters of each training batch;

[0028] S43. Normalize the training microseismic feature parameters using the mean and variance, and transform the training microseismic feature parameters into a standard normal distribution with a mean of 0 and a variance of 1 to obtain a 0 - 1 distribution;

[0029] S44. Perform scale transformation and offset on the normalized microseismic feature parameters.

[0030] Furthermore, the mean calculation of the microseismic feature parameters in step S41 adopts the following formula:

[0031]

[0032] where μ B is the mean of the training batch microseismic feature parameters, M is the number of training batch microseismic feature parameters, and x i is the i-th training microseismic feature parameter.

[0033] Furthermore, the variance calculation of the microseismic feature parameters in step S42 adopts the following formula:

[0034]

[0035] Among them, is the variance of the microseismic feature parameters of the training batch, M is the number of microseismic feature parameters of the training batch, and x i is the i-th training microseismic feature parameter, and μ B is the mean of the microseismic feature parameters of the training batch.

[0036] Furthermore, the normalization calculation of the microseismic feature parameters in step S43 adopts the following formula:

[0037]

[0038] Among them, x i is the i-th training microseismic feature parameter, μ B is the mean of the microseismic feature parameters of the training batch, is the variance of the microseismic feature parameters of the training batch, and ε is a small positive number used to avoid division by zero.

[0039] Furthermore, the scale transformation and offset calculation of the microseismic feature parameters in step S44 adopt the following formula:

[0040]

[0041] Among them, γ is the scale factor, x i is the i-th training microseismic feature parameter, and β is the translation factor.

[0042] Compared with the prior art, the intelligent rock failure mode discrimination method based on a unique microseismic signal provided by the present invention has the following advantages:

[0043] 1. The present invention utilizes the big data analysis ability of machine learning to realize the intelligent discrimination of the microseismic signal patterns of mixed rock failures, reducing the subjective error of dynamic disaster monitoring and early warning to a certain extent;

[0044] 2. In the process of realizing the discrimination of rock failure modes, the present invention only uses a single microseismic signal to complete the discrimination of rock failure modes, without using the variation law of microseismic signals in continuous time, reducing the objective error of dynamic disaster monitoring and early warning caused by reasons such as sensor failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a schematic flow chart of the intelligent rock failure mode discrimination method based on a unique microseismic signal provided by the present invention.

[0046] Figure 2 is a schematic diagram of the original microseismic waveform provided by the present invention.

[0047] Figure 3 is a schematic diagram of the extraction of the characteristic parameters of the microseismic waveform in the time domain and frequency domain provided by the present invention.

[0048] Figure 4 It is the one-dimensional feature map of microseismic waveform and microseismic parameters provided by the present invention.

[0049] Figure 5 It is a schematic structural diagram of the rock failure mode discrimination network M_Net provided by the present invention.

[0050] Figure 6 It is a schematic diagram of the performance change of the intelligent rock failure mode discrimination model provided by the present invention.

[0051] Figure 7 It is the discrimination result of the new method provided by the present invention for three types of failure modes in mines. Detailed implementation manners

[0052] In order to make the technical means, creative features, achieved purposes and functions of the present invention easy to understand, the present invention will be further described below with reference to specific drawings.

[0053] Please refer to Figure 1 As shown, the present invention provides an intelligent rock failure mode discrimination method based on unique microseismic signals, including the following steps:

[0054] S1. Collect the original microseismic waveforms of known rock failure modes; the schematic diagram of the original microseismic waveforms is as Figure 2 shown, where the horizontal axis is the time axis and the vertical axis is the energy amplitude information.

[0055] S2. Use Fourier transform to extract the characteristic parameters in the time domain and frequency domain of the microseismic waveform, that is, perform parameter extraction on the microseismic waveform through Fourier transform, and calculate the characteristic parameters in the time domain and frequency domain of the microseismic waveform. The parameter extraction diagram is as Figure 3 shown; the characteristic parameters include amplitude, duration, rise time, ring count, rise count, energy, effective voltage, average level, centroid frequency domain, peak frequency.

[0056] S3. Perform deviation standardization on the characteristic parameters in the time domain and frequency domain of the extracted microseismic waveform, and tile the standardized data of the characteristic parameters into a one-dimensional feature map of microseismic parameters. The length of the feature map is 10 and the height is 1; the one-dimensional feature map of microseismic parameters is as Figure 4 shown, where x1 to x 10 are the normalized values of the microseismic characteristic parameters.

[0057] S4. Construct a rock failure mode discrimination network M_Net with a convolutional neural network as the basic framework.

[0058] S5. Take the one-dimensional feature map of microseismic parameters as the input feature, the known rock failure modes as the labels, and the rock failure mode discrimination network M_Net as the basic framework. Train and learn the mapping relationship between the one-dimensional feature map of microseismic parameters and the discrimination of rock failure modes, and construct an intelligent discrimination model for rock failure modes.

[0059] S6. Verify the intelligent discrimination model for rock failure modes: Based on the discrimination accuracy of rock failure modes, when the accuracy is less than the preset value, adjust the hyperparameters of the intelligent discrimination model for rock failure modes, and retrain the model according to step S5 until the accuracy is greater than or equal to the preset value. Then it is considered that the model has been trained well and can be applied. As an implementation method, the preset value is set to 90%.

[0060] S7. Process the microseismic waveforms with unknown failure modes according to steps S2 and S3 above, and then input them into the trained intelligent discrimination model for rock failure modes in step S5 for failure mode discrimination, so as to realize the application of microseismic waveform discrimination for unknown rock failure modes.

[0061] As a specific embodiment, the Fourier transform in step S2 adopts the following formula:

[0062]

[0063] where F(ω) represents the spectrum after Fourier transform, f(t) is the original signal, e -iωt is the complex exponential function, and ω is the frequency.

[0064] As a specific embodiment, the deviation normalization in step S3 adopts the following formula:

[0065]

[0066] where x new is the result after normalization, x is the data to be normalized, x max is the maximum value of the same type of eigenvalue data, and x min is the minimum value of the same type of eigenvalue data.

[0067] As a specific embodiment, please refer to Figure 5 As shown, the rock failure mode discrimination network M_Net constructed in step S4 includes four convolutional blocks connected in sequence. Each convolutional block includes a convolutional layer (Conv), a batch normalization layer (Batchnorm), an activation layer (Relu), and a max pooling layer (Maxpool) connected in sequence from the input end to the output end. The convolutional kernel size of the convolutional layer of each convolutional block is 1×1. The number of convolutional kernels of the convolutional layers in the four convolutional blocks is 32, 64, 128, and 256 in sequence. The pooling kernel size of the max pooling layer in the four convolutional blocks is 1×1, and the pooling stride is 1.

[0068] As a specific embodiment, the convolution calculation formula of the convolution layer is as follows:

[0069]

[0070] Among them, I9i - m,j - n) is the input signal, K(m,n) is the convolution kernel, i,j are the indices of the output feature map, and m,n are the indices of the convolution kernel.

[0071] As a specific embodiment, the implementation of the batch normalization layer includes the following steps:

[0072] S41. Calculate the mean value of the microseismic feature parameters of each training batch;

[0073] S42. Calculate the variance of the microseismic feature parameters of each training batch;

[0074] S43. Normalize the training microseismic feature parameters of this batch using the mean value and variance, and transform the training microseismic feature parameters into a standard normal distribution with a mean of 0 and a variance of 1 to obtain a 0 - 1 distribution;

[0075] S44. Perform scale transformation and offset on the normalized microseismic feature parameters.

[0076] By adopting the implementation steps of the batch normalization layer in this embodiment, the convergence speed of the rock failure mode discrimination model training can be accelerated, and the rock failure mode discrimination model is easier to train and more stable.

[0077] As a specific embodiment, the mean value calculation of the microseismic feature parameters in step S41 adopts the following formula:

[0078]

[0079] Among them, μ B is the mean value of the microseismic feature parameters of the training batch, M is the number of microseismic feature parameters of the training batch, and x i is the i - th training microseismic feature parameter.

[0080] As a specific embodiment, the variance calculation of the microseismic feature parameters in step S42 adopts the following formula:

[0081]

[0082] Among them, is the variance of the microseismic feature parameters of the training batch, M is the number of microseismic feature parameters of the training batch, x i is the i - th training microseismic feature parameter, and μ B is the mean value of the microseismic feature parameters of the training batch.

[0083] As a specific embodiment, the normalization calculation of the microseismic characteristic parameters in step S43 adopts the following formula:

[0084]

[0085] where x i is the i-th training microseismic characteristic parameter, μ B is the mean value of the microseismic characteristic parameters in the training batch, is the variance of the microseismic characteristic parameters in the training batch, and ε is a small positive number used to avoid division by zero.

[0086] As a specific embodiment, the scale transformation and offset calculation of the microseismic characteristic parameters in step S44 adopt the following formula:

[0087]

[0088] where γ is the scale factor, x i is the i-th training microseismic characteristic parameter, and β is the translation factor.

[0089] As a specific embodiment, the activation function of the activation layer selects the ReLU function. The rectified linear unit ReLU is used as the non-linear excitation function to perform non-linear transformation on each normalized value. The rectified linear unit ReLU is defined as follows:

[0090] ReLU(y i ) = max(0, y i )

[0091] where ReLU(y i ) represents the rectified linear unit function, max represents finding the maximum value, and y i is an input value.

[0092] As a specific embodiment, Figure 6 During the process of mapping the one-dimensional feature map of the training and learning microseismic parameters to the distinction of rock failure modes, the performance of the intelligent rock failure mode distinction model changes. The final distinction accuracy of the model can reach 99.9%.

[0093] As a specific embodiment, Figure 7 For the distinction results of the three types (compression, shear, and tension) of failure modes in the mine by this method, the accuracy rate is above 99%, which proves the feasibility of the new method proposed by the present invention in the distinction of rock failure modes.

[0094] Compared with the prior art, the intelligent rock failure mode distinction method based on the unique microseismic signal provided by the present invention has the following advantages:

[0095] 1. The present invention utilizes the machine learning big data analysis ability to achieve the intelligent distinction of the microseismic signal patterns of the mixed rock failure, reducing the subjective error of the dynamic disaster monitoring and early warning to a certain extent.

[0096] 2. In the process of realizing the distinction of the rock failure mode, the present invention only uses a single microseismic signal to complete the distinction of the rock failure mode, without using the variation law of the microseismic signal in continuous time, reducing the objective error of the dynamic disaster monitoring and early warning caused by reasons such as sensor failure.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. An intelligent discrimination method for rock failure modes based on unique microseismic signals, characterized in that It includes the following steps: S1. Collect the original microseismic waveforms of known rock failure modes; S2. Use Fourier transform to extract the characteristic parameters in the time domain and frequency domain of the microseismic waveforms. The characteristic parameters include amplitude, duration, rise time, ring count, rise count, energy, effective voltage, average level, centroid frequency domain, and peak frequency; S3. Perform deviation normalization on the characteristic parameters in the time domain and frequency domain of the extracted microseismic waveforms, and tile the normalized data of the characteristic parameters into a one-dimensional microseismic parameter feature map; S4. With a convolutional neural network as the basic framework, construct a rock failure mode discrimination network M_Net; S5. Using the one-dimensional microseismic parameter feature map as the input feature, the known rock failure modes as labels, and the rock failure mode discrimination network M_Net as the basic framework, train and learn the mapping relationship between the one-dimensional microseismic parameter feature map and rock failure mode discrimination, and construct an intelligent rock failure mode discrimination model; S6. Verify the intelligent rock failure mode discrimination model: Based on the accuracy of rock failure mode discrimination, when the accuracy is less than the preset value, adjust the hyperparameters of the intelligent rock failure mode discrimination model, and retrain the model according to step S5 until the accuracy is greater than or equal to the preset value, then it is considered that the model has been trained well and can be applied; S7. Process the microseismic waveforms of unknown failure modes according to steps S2 and S3 above, and then input them into the trained intelligent rock failure mode discrimination model in step S5 for failure mode discrimination, to achieve the application of distinguishing microseismic waveforms of unknown rock failure modes.

2. The intelligent discrimination method for rock failure modes based on unique microseismic signals according to claim 1, wherein The Fourier transform in step S2 uses the following formula: Among them, F(ω) represents the spectrum after Fourier transform, f(t) is the original signal, and e -iωt is a complex exponential function, and ω is the frequency.

3. The intelligent discrimination method for rock failure modes based on unique microseismic signals according to claim 1, characterized in that The deviation normalization in step S3 uses the following formula: where x new is the result after standardization, x is the data to be standardized, and x max is the maximum value of the same type of eigenvalue data, and x min is the maximum value of the same type of eigenvalue data.

4. The intelligent discrimination method for rock failure modes based on unique microseismic signals according to claim 1, characterized in that The constructed rock failure mode discrimination network M_Net in step S4 includes four convolutional blocks connected in sequence. Each convolutional block includes a convolutional layer, a batch normalization layer, an activation layer, and a max pooling layer connected in sequence from the input end to the output end. The convolutional kernel size of the convolutional layer in each convolutional block is 1×1. The number of convolutional kernels in the convolutional layers of the four convolutional blocks is 32, 64, 128, and 256 in sequence. The pooling kernel size of the max pooling layer in the four convolutional blocks is 1×1, and the pooling stride is 1.

5. The intelligent discrimination method for rock failure modes based on unique microseismic signals according to claim 4, characterized in that, The convolution calculation formula of the convolutional layer is as follows: Among them, I(i - m, j - n) is the input signal, K(m, n) is the convolutional kernel, i, j are the indices of the output feature map, and m, n are the indices of the convolutional kernel.

6. The intelligent discrimination method for rock failure modes based on unique microseismic signals according to claim 4, characterized in that The implementation of the batch normalization layer includes the following steps: S41. Calculate the mean of the microseismic characteristic parameters of each training batch; S42. Calculate the variance of the microseismic characteristic parameters of each training batch; S43. Use the mean and variance to normalize the training microseismic characteristic parameters of this batch, and transform the training microseismic characteristic parameters into a standard normal distribution with a mean of 0 and a variance of 1 to obtain a 0 - 1 distribution; S44. Perform scale transformation and offset on the normalized microseismic characteristic parameters.

7. The intelligent discrimination method for rock failure modes based on unique microseismic signals according to claim 6, characterized in that, The formula for calculating the mean of the microseismic characteristic parameters in step S41 is as follows: Among them, μ B is the mean value of the microseismic feature parameters of the training batch, M is the number of the microseismic feature parameters of the training batch, and x i is the i-th training microseismic feature parameter.

8. The intelligent discrimination method for rock failure modes based on unique microseismic signals according to claim 6, characterized in that, The formula for calculating the variance of the microseismic characteristic parameters in step S42 is as follows: Among them, is the variance of the microseismic feature parameters in the training batch, M is the number of microseismic feature parameters in the training batch, and x i is the i-th training microseismic feature parameter, and μ B is the mean of the microseismic feature parameters in the training batch.

9. The intelligent discrimination method for rock failure modes based on unique microseismic signals according to claim 6, characterized in that, The formula for calculating the normalization of the microseismic characteristic parameters in step S43 is as follows: Among them, x i is the i-th training microseismic feature parameter, μ B is the mean of the training batch microseismic feature parameters, is the variance of the training batch microseismic feature parameters, and ε is a small positive number used to avoid division by zero.

10. The intelligent discrimination method for rock failure modes based on unique microseismic signals according to claim 6, characterized in that In step S44, the scale transformation and offset calculation of the microseismic characteristic parameters adopt the following formula: where γ is the scale factor, x i is the i-th training microseismic feature parameter, and β is the translation factor.

Images