Microseismic detection method and device and readable medium

Abstract

The invention discloses a micro-seismic detection method and device and a readable medium, and the method comprises the steps: processing sample data, and obtaining noise reduction waveform data; establishing a deep convolutional neural network classification model, and training the deep convolutional neural network classification model by using the de-noised waveform data; establishing a support vector machine model, and training the support vector machine model by using the comprehensive feature vector; and inputting the processed to-be-detected data into the deep convolutional neural network classification model, and inputting the output of the deep convolutional neural network classification model into the support vector machine model to obtain a detection result. According to the invention, a mode of the deep convolutional neural network classification model and the support vector machine model is adopted, and the deep convolutional neural network classification model provides high-quality input data for the support vector machine model through the powerful feature extraction capability of the deep convolutional neural network classification model; and the support vector machine model performs final classification on the basis of the features extracted by the deep convolutional neural network classification model, so that the accuracy of the detection result is improved.

Description

Technical Field

[0001] The present invention relates to the field of microseismic detection, and in particular to a microseismic detection method, device and readable medium. Background Art

[0002] The recognition and classification of multi-channel microseismic waveforms is a key part of microseismic monitoring technology. Since microseismic signals are often weak and easily interfered by noise, how to accurately identify and classify these signals is crucial for engineering safety monitoring. At present, microseismic monitoring technology has been widely used in actual engineering application scenarios such as mining, tunneling and underground engineering, rail transportation, landslide and debris flow prevention, marine engineering, and nuclear waste disposal sites. Using existing labeled data, the classification model is trained through machine learning algorithms (such as support vector machines, neural networks, etc.). The model learns how to classify different seismic events based on the extracted features. The trained model is applied to newly collected signals to identify and classify seismic events in real time. The coordinated processing of data from multiple channels can improve the accuracy of classification. The classified results are used to analyze seismic activities in engineering scenarios, assess potential safety risks, and help engineering managers make decisions, such as preventing rock bursts and monitoring tunnel stability.

[0003] The process of multi-channel microseismic waveform recognition and classification generally includes signal acquisition, preprocessing, feature extraction, classification model training, real-time recognition and classification, and result analysis. Microseismic signals are collected in real time through multiple sensors deployed in engineering scenes such as mines and tunnels. These signals come from different positions and angles and can provide rich spatial information.

[0004] The existing technology has the following shortcomings: insufficient generalization ability of the model. Many existing models perform well on specific data sets, but when processing data in different mines or other scenarios such as rail transit safety, the performance may be unstable and the generalization ability is poor; noise and signal confusion. In complex actual engineering environments, microseismic signals are easily confused with noise and other types of signals (such as blasting signals, mechanical vibration signals), resulting in misclassification or missed detection; poor adaptability to dynamically changing environments. When faced with dynamically changing environments, many methods find it difficult to quickly adapt to newly emerging signal patterns, resulting in a decrease in classification accuracy.

[0005] The main reasons for these limitations and shortcomings include the complexity and diversity of microseismic signals and the limitations of existing technologies in manual feature extraction and reliance on specific scenarios. Future development may require more automated and intelligent algorithms that can better handle complex signal features, adapt to different application scenarios, and improve classification accuracy and generalization while reducing manual intervention. Summary of the invention

[0006] In order to overcome the above technical defects, the present invention provides a microseismic detection method, device and readable medium, which can improve the accuracy of detection.

[0007] In order to solve the above problems, the present invention is implemented according to the following technical solutions:

[0008] A microseismic detection method comprises the steps of:

[0009] Processing the sample data to obtain noise-reduced waveform data and standardized feature vectors;

[0010] A deep convolutional neural network classification model was built and trained using denoised waveform data;

[0011] Establish a support vector machine model and train it using comprehensive feature vectors;

[0012] The data to be tested is input into the deep convolutional neural network classification model, and the output of the deep convolutional neural network classification model is input into the support vector machine model to obtain the detection result.

[0013] As a further improvement of the present invention, the step of filtering the sample data to obtain the noise-reduced waveform data includes:

[0014] The sample data is filtered using the wavelet threshold denoising method to obtain the denoised waveform data;

[0015] The Z-score standardization method is used to standardize the sample data and obtain the standardized feature vector.

[0016] As a further improvement of the present invention, in the step of establishing a deep convolutional neural network classification model and training it with denoised waveform data, the denoised waveform data is converted into two-dimensional image data and mapped into a time-amplitude matrix as input to the deep convolutional neural network classification model.

[0017] As a further improvement of the present invention, the deep convolutional neural network classification model includes: multiple convolutional layers, pooling layers, and fully connected layers. In the convolutional layers, the first convolutional layer uses a 7×7 convolution kernel, and then uses a 3×3 and 5×5 mixed convolution kernel.

[0018] As a further improvement of the present invention, the step of using the noise reduction waveform data to train it includes:

[0019] Use cross entropy as the loss function;

[0020] The back-propagation algorithm is used to calculate the gradient of the loss function to the weights of each layer, and the weights of the deep convolutional neural network classification model are updated based on the weights of each layer;

[0021] The Adam optimization algorithm is used to update the parameters of the deep convolutional neural network classification model;

[0022] Regularization methods are used to enhance the generalization ability of deep convolutional neural network classification models.

[0023] As a further improvement of the present invention, the output of the deep convolutional neural network classification model is: a feature vector for each channel;

[0024] Before the step of inputting the output of the deep convolutional neural network classification model into the support vector machine model, the method further includes:

[0025] The feature vectors of all channels are combined to obtain a comprehensive feature vector used to characterize the comprehensive features of all channels.

[0026] As a further improvement of the present invention, in the step of establishing the support vector machine model and training it with sample data, a radial basis function kernel is used.

[0027] As a further improvement of the present invention, the step of establishing a support vector machine model and training it using sample data includes:

[0028] Based on different numbers of channels, different support vector machine models are trained.

[0029] The present invention also provides an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0030] The memory is used to store computer programs;

[0031] The processor is used to implement the above-mentioned microseismic detection method when executing the program stored in the memory.

[0032] The present invention also provides one or more computer-readable media having instructions stored thereon, which, when executed by one or more processors, enable the processors to perform the above-mentioned microseismic detection method.

[0033] Compared with the prior art, the present invention has the following beneficial effects: the deep convolutional neural network classification model is responsible for extracting multi-channel features from each sensor and inputting them into the support vector machine model. The support vector machine model can realize efficient classification of multi-channel data and improve classification accuracy. By adopting the deep convolutional neural network classification model and the support vector machine model, the deep convolutional neural network classification model provides high-quality input data for the support vector machine model through its powerful feature extraction capability, and the support vector machine model performs the final classification based on the features extracted by the deep convolutional neural network classification model, thereby improving the accuracy of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings, wherein:

[0035] Figure 1 This is a flow chart of the microseismic detection method described in Example 1. DETAILED DESCRIPTION

[0036] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0037] Example 1

[0038] The present invention provides a microseismic detection method, such as Figure 1 As shown, the steps include:

[0039] S1. Collect sample data as training set and test set; the sample data is divided into two categories: the first category of sample data consists of single waveform images, including microseismic images, explosion images and noise images, each image can have 10,000 samples, and a total of 30,000 samples are used to train and test the deep convolutional neural network classification model; the second category of sample data consists of single event multi-channel waveform images, each image contains multiple waveform images, including explosion events, microseismic events and noise events. Each event has 1,000 samples, and a total of 3,000 event samples are used to train and test the SVM classifier. In addition, the waveform images of the first data set are marked; 0 represents microseismic images, 1 represents explosion images, and 2 represents noise images. Finally, the data in the sample data are put into the training set and test set at a ratio of 3:1.

[0040] S2. Process the sample data to obtain noise-reduced waveform data and standardized feature vectors. Due to complex underground environmental conditions, microseismic signals are severely interfered with during transmission. Interference signals include motor vehicle noise, machinery, equipment noise, and electromagnetic interference. Most monitoring data are composed of various valid signals and interference signals, so signal preprocessing must be performed first, including filtering, denoising and other steps to remove background noise and retain valid seismic signals. Key features such as amplitude, frequency, duration, etc. are extracted from the preprocessed signal. These features can help distinguish different types of seismic events (such as natural earthquakes and artificially induced earthquakes). In order to more accurately extract the basic features of the waveform image, this embodiment first uses the wavelet threshold denoising method to denoise and filter the first type of sample data.

[0041] In the process of denoising and filtering, the amplitude of the effective signal does not change with the increase of the wavelet analysis scale, but the interference signal can be quickly attenuated to eliminate the redundant interference signal in the effective signal. For the one-dimensional original signal model of the noise signal contained in formula (1).

[0042] s(k)=f(k)+εe(k) (k=0,1,…,n-1) (1)

[0043] Where s(k), f(k), ε and e(k) are the signal containing noise, the low-frequency stationary signal, the noise amplification factor and Gaussian white noise respectively.

[0044] First, the first type of sample data s(k) is subjected to wavelet transform to obtain coefficients of multiple scales:

[0045]

[0046] Among them, A j is the low frequency coefficient (low frequency signal), D j It is the high frequency coefficient (noise part). The coefficients of different scales contain different frequency information of the signal.

[0047] Then, threshold processing is applied to the high-frequency coefficients. Since noise is generally concentrated in the high-frequency part, a threshold T is set and the following operations are performed on the high-frequency coefficients:

[0048] D' j =Threshold(D j ,T) (3)

[0049] Threshold function represents the j The coefficients are thresholded. Taking the soft threshold method as an example, the specific function is:

[0050] D' j =sign(D j )·max(|D j |-T,0) (4)

[0051] In this way, the coefficients below the threshold are set to zero, retaining important high-frequency information, thereby removing noise. The denoised signal is then reconstructed through the inverse wavelet transform:

[0052]

[0053] The denoised coefficients are combined into the final denoised signal through inverse transformation. The wavelet threshold denoising method is used to denoise the first type of sample data, thereby greatly increasing the total amount of data that can be used for deep convolutional neural network training, so as to improve the generalization ability of the deep convolutional neural network classification model training weight model in subsequent steps and prevent overfitting. In addition, the wavelet threshold denoising method can also enhance the effect of effective signals, suppress the influence of noise interference, and more accurately extract the basic features of waveform image data.

[0054] Wavelet threshold denoising can not only double the amount of sample data in the first category, ensuring that the deep convolutional neural network classification model has enough data for training and testing, but also obtain high-quality waveform images, thereby greatly suppressing noise interference. This effect makes the effective information input into the deep convolutional neural network classification model greater than the useless information, and the learned features are easier to identify and classify. At the same time, the original image without wavelet threshold denoising and filtering in the first category of sample data can prevent the loss of effective information of the filtered image, thereby solving the problem of incomplete learning features of the deep convolutional neural network classification model.

[0055] In order to increase the randomness of samples, improve the generalization ability of the deep convolutional neural network classification model, and make feature extraction more accurate, the denoised and labeled waveform images are randomly divided into 10 images. Instead of directly performing forward calculations on the images, forward operations are performed on the 10 cropped images formed by the first type of sample data to extract features, and the feature mean of the 10 cropped images extracted by the deep convolutional neural network classification model is obtained. The maximum number of iterations of the deep convolutional neural network classification model is set, the appropriate learning rate during training is set, and the appropriate momentum factor is set to make the weight update smooth, stable, and fast. In addition, all training sample values ​​are subtracted from their average value to improve and speed up the convergence speed.

[0056] S3. Establish a deep convolutional neural network classification model and train it using denoised waveform data.

[0057] Prior to this, the denoised waveform data needs to be further processed to convert it into two-dimensional image data and map it into a time-amplitude matrix. Each microseismic event is recorded by multiple sensors (channels), and the denoised waveform data of each channel is regarded as a channel of the image. These two-dimensional image data are used as input to the deep convolutional neural network classification model.

[0058] The deep convolutional neural network classification model is used for forward propagation training, which includes: multiple convolutional layers, pooling layers, and fully connected layers:

[0059] (1) Convolutional layer: The convolutional layer extracts the features of the input feature map (or image) through convolution operations. By filtering the waveform image, low-level and high-level features (such as amplitude, frequency, etc.) are gradually extracted.

[0060] In this embodiment, the first convolution layer of the deep convolutional neural network classification model uses a 7×7 convolution kernel, and the next convolution layer uses a 3×3 and 5×5 mixed convolution kernel to extract the features of the input waveform image. The mixed convolution kernel can extract features of different scales and reduce the connection parameters between neurons. The convolution kernel is randomly initialized and optimized by the back-propagation algorithm. The output of the convolution layer is defined as:

[0061]

[0062] Relu represents the activation function, represents the i-th input of the j-th neuron in the l-th layer, represents the size of the convolution kernel between neuron j in layer l and neuron i in layer l-1, * represents the convolution operation, Mj represents the selection of the input graph, is the additive bias of neuron j in layer l.

[0063] (2) Pooling layer: The pooling operation reduces the resolution of the feature map but retains the most important information, thereby enhancing the model's ability to recognize waveform features.

[0064] In order to reduce the amount of computation and improve the robustness of the model, the deep convolutional neural network classification model uses maximum pooling and average pooling layers. The output of the pooling layer is defined as:

[0065]

[0066] In formula (7), pooling() represents the maximum or average pooling function. β is a multiplicative bias. The activation function Relu can introduce nonlinearity, allowing the deep convolutional neural network classification model to learn complex feature patterns.

[0067] (3) Fully connected layer: In order to reduce the number of parameters and avoid overfitting, the traditional fully connected layer is replaced by a global average pooling layer.

[0068] The global average pooling layer converts the multi-dimensional feature map into a one-dimensional vector, significantly reducing the parameters of the deep convolutional neural network classification model and improving computational efficiency.

[0069] (4) Output layer: The output of the deep convolutional neural network classification model is converted into classification probability through the Softmax function, indicating which type of microseismic event the input waveform belongs to.

[0070] Based on the denoising of the original waveform to obtain the denoised waveform data, a deep convolutional neural network classification model is used to automatically extract waveform features to identify and classify single-event multi-channel microseismic events.

[0071] Training and optimizing a deep convolutional neural network classification model using denoised waveform data includes the following steps:

[0072] (1) Forward propagation: The deep convolutional neural network classification model built through steps gradually extracts and abstracts features through operations such as convolution, pooling, and activation.

[0073] (2) Loss function calculation: The cross entropy loss function is used to measure the difference between the network's prediction and the true label. The cross entropy loss function is defined as:

[0074]

[0075] In formula (8), N represents the number of categories, y i is the actual type label, is the predicted category probability, given by the Softmax function. The role of the cross entropy loss function is to minimize the difference between the model output and the true label, making the model's predicted value closer to the true value.

[0076] (3) Back propagation: The back propagation algorithm calculates the gradient of the loss function with respect to the weights of each layer and updates the weights of the deep convolutional neural network classification model. Through the chain rule, the error is propagated back from the output of the network to each layer to minimize the loss function. The specific steps of back propagation include:

[0077] The output layer gradient is calculated using the derivative of the cross entropy loss function:

[0078]

[0079] Using the chain rule, back propagate the gradient of each layer and calculate the hidden layer gradient. For the convolution layer, assuming that the output of the deep convolutional neural network classification model is y and the weight is w, then the hidden layer gradient is:

[0080]

[0081] Use gradient descent to update parameters:

[0082]

[0083] Where η is the learning rate, which determines the step size of each update.

[0084] (4) Optimization algorithm: The Adam optimization algorithm is used to update the parameters of the deep convolutional neural network classification model to make the loss function as small as possible, ensure that the deep convolutional neural network classification model gradually converges and improves the classification accuracy. The update rules of the commonly used adaptive Adam optimization algorithm are as follows:

[0085]

[0086] in and are the momentum term and the square of the gradient, respectively, β1 and β2 are hyperparameters, and ε is a constant to prevent division by zero errors.

[0087] (5) Regularization technology: During the training process, regularization techniques such as Dropout and batch normalization are used to reduce overfitting and enhance the generalization ability of the network. The Dropout formula is as follows:

[0088]

[0089] Where x is the neuron output and p is the drop probability, which is usually set to 0.5. During the training process, Dropout randomly drops some neurons, making the structure of the deep convolutional neural network classification model different each time it is forward propagated, thereby reducing the risk of overfitting. Batch normalization standardizes the input of each layer so that the mean of the input of each layer is 0 and the variance is 1, which helps to speed up training and reduce the problem of vanishing gradients.

[0090] After the deep convolutional neural network classification model is trained, it can capture complex patterns and details in waveform signals through multi-layer convolution, and can be used to extract features from multi-channel microseismic waveforms. The deep convolutional neural network classification model is mainly used for automatic identification and classification of multi-channel microseismic events.

[0091] After the denoised and filtered microseismic signal image to be detected is input into the constructed deep convolutional neural network classification model to extract features, the feature vector of each channel is output, and the feature vectors of all channels are merged to obtain a comprehensive feature vector used to characterize the comprehensive features of all channels; since the number of channels for each event (that is, the dimension of each input data) may be different, resulting in changes in the final feature dimension, it is necessary to first standardize the feature vectors to ensure that all data input to the support vector machine model have the same scale, and feature vectors with different numbers of channels can be used as inputs to the support vector machine model. Standardization usually uses the Z-score standardization method:

[0092]

[0093] where x i is the original feature, μ is the feature mean, and σ is the standard deviation. This process scales each feature value to a range of mean 0 and standard deviation 1, ensuring the uniformity and stability of the data.

[0094] S4. Establish a support vector machine model and train it using sample data.

[0095] The performance of the support vector machine model often depends on the choice of kernel function. In this embodiment, the radial basis function kernel (RBF) is selected because it can handle nonlinear separable problems and performs well in high-dimensional space. The formula of the RBF kernel function is as follows:

[0096]

[0097] where x i and x j are the feature vectors of the two samples, ||x i -x j|| is the Euclidean distance between samples, and σ is the parameter of the RBF kernel function, which controls the width of the kernel function.

[0098] The training of the support vector machine model is to learn the optimal classification hyperplane by training the features of each event. By maximizing the boundary between categories, the support vector machine model ensures that the model has better classification performance. In high-dimensional space, the support vector machine model finds the optimal hyperplane by solving an optimization problem. The specific optimization goal is:

[0099]

[0100] Where ξ, w, and C represent slack variables, weight vectors, and penalty coefficients, respectively, indicating the degree of penalty for the support vector machine model. Applying the Lagrange multiplier method to solve the optimal classification hyperplane can be transformed into the following constrained optimization problem:

[0101]

[0102] The constraint optimization is as follows:

[0103]

[0104] in represents the Lagrange multiplier, most a i The value of is 0, and the samples corresponding to the samples whose value is not 0 are called support vectors. In the training process of the support vector machine model, support vectors refer to training samples with non-zero Lagrange multipliers, which directly affect the determination of the classification hyperplane. The final classification decision function is:

[0105]

[0106] where α i is the coefficient of the support vector; y i is the label of the support vector; b is the bias term. Through this decision function, SVM can determine which category the input sample x belongs to and finally give the classification result.

[0107] Since the number of sensors for each event is different, different SVM models need to be trained for different numbers of channels (such as 5 channels, 7 channels, etc.). The input feature dimension of each SVM model is adjusted according to the number of channels to ensure that the dimension of the feature vector is consistent.

[0108] The trained support vector machine model can be used to make predictions based on new multi-channel waveform data. The feature vector of each multi-channel waveform is classified by the support vector machine model, and the classification results are finally output, such as event type, epicenter location, etc.

[0109] The deep convolutional neural network classification model is responsible for extracting high-dimensional features from the waveform image of each sensor (channel), while the support vector machine model uses these extracted features to make the final classification decision. By using the RBF kernel function and the constraint method of the optimization problem, the support vector machine model can effectively process features in high-dimensional space and provide high-precision classification results.

[0110] S5. Input the processed data to be detected into the deep convolutional neural network classification model, and input the output of the deep convolutional neural network classification model into the support vector machine model to obtain the detection result.

[0111] The output of the last convolutional layer or fully connected layer of the deep convolutional neural network classification model is used as a standardized feature vector, which represents the characteristics of the input signal. The standardized feature vector is input into the support vector machine model, which performs classification operations based on the input high-dimensional feature vector and outputs the prediction results. Each microseismic event will be classified into different categories (such as normal signals, blasting signals, rockburst signals, etc.). The final output is the classification result of each microseismic event, which can be used for real-time monitoring and early warning. Based on the output results, the system can respond in real time, such as issuing an early warning when a dangerous signal is detected, or recording the normal operating status. Based on the classification results, the deep convolutional neural network classification model and the support vector machine model can be post-processed and adjusted.

[0112] Example 2

[0113] This embodiment provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the microseismic detection method in Example 1 when executing the program stored in the memory.

[0114] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk SolidStateDisk (SSD)), etc.

[0115] Example 3

[0116] This embodiment provides one or more computer-readable media on which instructions are stored. When executed by one or more processors, the processors are enabled to perform the microseismic detection method in Embodiment 1.

[0117] Optionally, the computer-readable storage medium may include: a read-only memory (ROM), a random access memory (RAM), a solid state drive (SSD), or an optical disk.

[0118] Among them, the random access memory may include resistance random access memory (ReRAM, Resistance Random Access Memory) and dynamic random access memory (DRAM, Dynamic Random Access Memory).

[0119] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A microseismic detection method, characterized in that: Includes steps: Processing the sample data to obtain noise-reduced waveform data; A deep convolutional neural network classification model was built and trained using denoised waveform data; Establish a support vector machine model and train it using comprehensive feature vectors; The processed data to be tested is input into the deep convolutional neural network classification model, and the output of the deep convolutional neural network classification model is input into the support vector machine model to obtain the detection result.

2. The microseismic detection method according to claim 1, characterized in that: The step of filtering the sample data to obtain the noise-reduced waveform data comprises: The wavelet threshold denoising method is used to filter the sample data to obtain the denoised waveform data.

3. The microseismic detection method according to claim 1, characterized in that: In the step of establishing a deep convolutional neural network classification model and training it using denoised waveform data, the denoised waveform data is converted into two-dimensional image data and mapped into a time-amplitude matrix as input to the deep convolutional neural network classification model.

4. The microseismic detection method according to claim 1, characterized in that: The deep convolutional neural network classification model includes: multiple convolutional layers, pooling layers, and fully connected layers. In the convolutional layers, the first convolutional layer uses a 7×7 convolution kernel, and then uses a 3×3 and 5×5 mixed convolution kernel.

5. The microseismic detection method according to claim 4, characterized in that: The step of training the denoised waveform data comprises: Use cross entropy as the loss function; The back-propagation algorithm is used to calculate the gradient of the loss function to the weights of each layer, and the weights of the deep convolutional neural network classification model are updated based on the weights of each layer; The Adam optimization algorithm is used to update the parameters of the deep convolutional neural network classification model; Regularization methods are used to enhance the generalization ability of deep convolutional neural network classification models.

6. The microseismic detection method according to claim 1, characterized in that: The output of the deep convolutional neural network classification model is: the feature vector for each channel; Before the step of inputting the output of the deep convolutional neural network classification model into the support vector machine model, the method further includes: Merge the feature vectors of all channels to obtain a comprehensive feature vector for characterizing the comprehensive features of all channels; The Z-score standardization method is used to standardize the comprehensive feature vector to obtain the standardized feature vector.

7. The microseismic detection method according to claim 6, characterized in that: In the step of establishing a support vector machine model and training it with sample data, a radial basis function kernel is used.

8. The microseismic detection method according to claim 6, characterized in that: The step of establishing a support vector machine model and training it using sample data includes: Based on different numbers of channels, different support vector machine models are trained.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; The memory is used to store computer programs; The processor is used to implement the microseismic detection method according to any one of claims 1 to 8 when executing the program stored in the memory.

10. One or more computer-readable media having instructions stored thereon, which, when executed by one or more processors, enable the processors to perform the microseismic detection method according to any one of claims 1 to 8.

Images