A method and device for adaptive automatic classification of mine microseismic monitoring data

Abstract

The present invention relates to the technical field of signal classification, and discloses a method and device for adaptive automatic classification of mine microseismic monitoring data. The method comprises: acquiring mine microseismic monitoring data, performing short-time Fourier transform on the mine microseismic monitoring data to obtain time-frequency domain features, and extracting multi-scale time-frequency domain information of the time-frequency domain features; extracting frequency domain energy distribution features of mine microseismic monitoring data in the frequency domain; establishing a self-attention sequence model in the time domain, and extracting time domain sequence features of mine microseismic monitoring data according to the self-attention sequence model; calculating multi-sensor aggregation features of mine microseismic monitoring data according to frequency domain energy distribution features and time domain sequence features; using a multi-layer perceptron to classify multi-sensor aggregation features, and performing positive and negative sample constraints on multi-sensor aggregation features to obtain classification results of mine microseismic monitoring data. The present application can improve the timeliness of data processing, realize effective feature extraction and fusion, and improve the accuracy and reliability of classification results.

Description

Technical Field

[0001] The present invention relates to the technical field of signal classification, and in particular to a method and device for adaptive automatic classification of mine microseismic monitoring data. Background Art

[0002] The traditional adaptive automatic classification method of mine microseismic monitoring data uses peak recognition to classify microseismic data and operating blasting data, and relies on the time difference between different channels of data and the correlation in time and frequency domain to classify vibration data and operating electrical noise.

[0003] The above method is difficult to take into account the classification requirements of microseismic / blasting data under different working conditions at the same time. It requires manual adjustment of parameters to adapt to changes in working conditions and is limited by the accuracy of data picking at each channel. Summary of the invention

[0004] In view of this, the purpose of the present invention is to overcome the deficiencies in the prior art and to provide a method and device for adaptive automatic classification of mine microseismic monitoring data.

[0005] The present invention provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present disclosure provides a method for adaptive automatic classification of mine microseismic monitoring data, which is applied to an adaptive automatic classification model of mine microseismic monitoring data, and the method includes:

[0007] Acquire microseismic monitoring data of a mine, perform short-time Fourier transform on the microseismic monitoring data of the mine to obtain time-frequency domain features, and use a time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain features;

[0008] Multi-scale horizontal pooling is used to extract frequency domain energy distribution characteristics of the mine microseismic monitoring data in the frequency domain;

[0009] Establishing a self-attention sequence model in the time domain, and extracting the time domain sequence features of the mine microseismic monitoring data according to the self-attention sequence model;

[0010] Calculating multi-sensor aggregation features in the mine microseismic monitoring data according to the frequency domain energy distribution features and the time domain sequence features;

[0011] A multi-layer perceptron is used to classify the multi-sensor aggregate features, and a feature matching method is used to perform positive and negative sample constraints on the multi-sensor aggregate features to obtain the classification results of the mine microseismic monitoring data.

[0012] Optionally, performing short-time Fourier transform on the mine microseismic monitoring data to obtain time-frequency domain features, and using a time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain features, includes:

[0013] Obtaining a preset number of item parameters and a preset window length parameter, and performing a short-time Fourier transform on the mine microseismic monitoring data according to the preset number of item parameters and the preset window length parameter to obtain initial time-frequency domain features;

[0014] The positive frequency component in the initial time-frequency domain feature is used as the time-frequency domain feature, and the time-domain multi-scale convolutional network is used to extract multi-scale time-frequency domain information of the time-frequency domain feature.

[0015] Optionally, the extracting frequency domain energy distribution characteristics of the mine microseismic monitoring data in the frequency domain by using multi-scale horizontal pooling includes:

[0016] Performing multi-scale horizontal pooling on the multi-scale time-frequency domain information in the frequency dimension to obtain multiple down-sampled feature maps;

[0017] Each of the downsampled feature maps is spliced ​​along the frequency dimension to obtain the frequency domain energy distribution characteristics of the mine microseismic monitoring data.

[0018] Optionally, extracting the time domain sequence features of the mine microseismic monitoring data according to the self-attention sequence model includes:

[0019] Pooling the self-attention sequence model and the frequency domain energy distribution feature to obtain a self-attention sequence model pooling result and a frequency domain energy distribution feature pooling result of the self-attention sequence model and the frequency domain energy distribution feature in the final frequency dimension;

[0020] The pooling result of the self-attention sequence model is concatenated with the pooling result of the frequency domain energy distribution feature to obtain the time domain sequence characteristics of the mine microseismic monitoring data.

[0021] Optionally, the calculating, according to the frequency domain energy distribution feature and the time domain sequence feature, a multi-sensor aggregation feature in the mine microseismic monitoring data includes:

[0022] Performing multi-sensor feature extraction on the frequency domain energy distribution feature and the time domain sequence feature to obtain a feature vector;

[0023] The feature vector is pooled in the sensor dimension to obtain the multi-sensor aggregate feature.

[0024] Optionally, the multi-layer perceptron is used to classify the multi-sensor aggregate features, and a feature matching method is used to perform positive and negative sample constraints on the multi-sensor aggregate features to obtain the classification results of the mine microseismic monitoring data, including:

[0025] Preprocessing and constraining the multi-sensor aggregate features with positive and negative samples to obtain optimized multi-sensor aggregate features, and inputting the optimized multi-sensor aggregate features into the multi-layer perceptron for classification;

[0026] The output result of the multilayer perceptron is converted into a probability distribution using an activation function, and the classification result is obtained according to the probability distribution.

[0027] Optionally, the method for adaptive automatic classification of mine microseismic monitoring data further includes:

[0028] Obtaining a type label vector of the mine microseismic monitoring data, and performing short-time Fourier transform on the type label vector to obtain a label vector;

[0029] Calculating the cross entropy loss between the classification result and the label vector, and using the cross entropy loss as the feature classification loss;

[0030] Acquire the same category data and different category data of the mine microseismic monitoring data, calculate the triplet loss of the mine microseismic monitoring data, the same category data and the different category data based on the multi-scale time-frequency domain information, the frequency domain energy distribution characteristics, the time domain sequence characteristics and the multi-sensor aggregation characteristics, and use the triplet loss as the feature matching loss;

[0031] The adaptive automatic classification model of mine microseismic monitoring data is jointly constrained according to the feature classification loss and the feature matching loss.

[0032] In a second aspect, an embodiment of the present disclosure provides a device for adaptive automatic classification of microseismic monitoring data of a mine, which is applied to an adaptive automatic classification model of microseismic monitoring data of a mine, and the device includes:

[0033] A shallow feature extraction module is used to obtain mine microseismic monitoring data, perform short-time Fourier transform on the mine microseismic monitoring data to obtain time-frequency domain features, and use a time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain features;

[0034] A frequency domain energy distribution feature extraction module, used to extract the frequency domain energy distribution features of the mine microseismic monitoring data in the frequency domain by using multi-scale horizontal pooling;

[0035] A time domain sequence feature extraction module is used to establish a self-attention sequence model in the time domain, and to extract the time domain sequence features of the mine microseismic monitoring data according to the self-attention sequence model;

[0036] A multi-channel aggregation feature calculation module, used to calculate the multi-sensor aggregation features in the mine microseismic monitoring data according to the frequency domain energy distribution features and the time domain sequence features;

[0037] The classification module is used to classify the multi-sensor aggregation features using a multi-layer perceptron, and use a feature matching method to perform positive and negative sample constraints on the multi-sensor aggregation features to obtain the classification results of the mine microseismic monitoring data.

[0038] Beneficial effects of this application:

[0039] The method for adaptive automatic classification of mine microseismic monitoring data provided by the embodiment of the present application includes: acquiring mine microseismic monitoring data, performing short-time Fourier transform on the mine microseismic monitoring data to obtain time-frequency domain features, and using a time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain features; using multi-scale horizontal pooling to extract the frequency domain energy distribution features of the mine microseismic monitoring data in the frequency domain; establishing a self-attention sequence model in the time domain, and extracting the time domain sequence features of the mine microseismic monitoring data according to the self-attention sequence model; calculating the multi-sensor aggregation features in the mine microseismic monitoring data according to the frequency domain energy distribution features and the time domain sequence features; using a multi-layer perceptron to classify the multi-sensor aggregation features, and using a feature matching method to constrain the multi-sensor aggregation features with positive and negative samples to obtain the classification results of the mine microseismic monitoring data. The present application can improve the timeliness and accuracy of data processing, realize the effective extraction and fusion of features, and improve the accuracy and reliability of classification results.

[0040] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work. In each of the drawings, similar components are numbered similarly.

[0042] Figure 1 One of the flow charts of a method for adaptive automatic classification of mine microseismic monitoring data provided by an embodiment of the present application is shown;

[0043] Figure 2 A second flowchart of a method for adaptive automatic classification of mine microseismic monitoring data provided by an embodiment of the present application is shown;

[0044] Figure 3 A third flowchart of a method for adaptive automatic classification of mine microseismic monitoring data provided by an embodiment of the present application is shown;

[0045] Figure 4 A fourth flowchart of a method for adaptive automatic classification of mine microseismic monitoring data provided by an embodiment of the present application is shown;

[0046] Figure 5 A fifth flowchart of a method for adaptive automatic classification of mine microseismic monitoring data provided by an embodiment of the present application is shown;

[0047] Figure 6 A sixth flowchart of a method for adaptive automatic classification of mine microseismic monitoring data provided by an embodiment of the present application is shown;

[0048] Figure 7 A flowchart of a positive and negative sample constraint method provided in an embodiment of the present application is shown;

[0049] Figure 8 FIG7 shows a flowchart of a method for adaptive automatic classification of mine microseismic monitoring data provided by an embodiment of the present application;

[0050] Fig. 9 A schematic structural diagram of a device for adaptively and automatically classifying microseismic monitoring data of a mine provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0051] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0052] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. In contrast, when an element is referred to as being "directly on" another element, there is no intermediate element. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0053] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0054] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of the template are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0056] Example 1

[0057] like Figure 1 As shown, it is a flowchart of a method for adaptive automatic classification of mine microseismic monitoring data in an embodiment of the present application. The method for adaptive automatic classification of mine microseismic monitoring data provided in the embodiment of the present application is applied to an adaptive automatic classification model of mine microseismic monitoring data, and specifically includes the following steps:

[0058] Step S110, acquiring mine microseismic monitoring data, performing short-time Fourier transform on the mine microseismic monitoring data to obtain time-frequency domain features, and using a time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain features.

[0059] In this embodiment, the original mine microseismic monitoring data is first obtained. The mine microseismic monitoring data is a three-dimensional tensor x∈R n×m×l , where n is the number of axes, m is the number of sensors, and l is the number of sampling points.

[0060] In an optional embodiment, if Figure 2 As shown, step S110 includes:

[0061] Step S111, obtaining a preset number of items parameter and a preset window length parameter, and performing a short-time Fourier transform on the mine microseismic monitoring data according to the preset number of items parameter and the preset window length parameter to obtain initial time-frequency domain features;

[0062] Step S112: taking the positive frequency component in the initial time-frequency domain feature as the time-frequency domain feature, and using the time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain feature.

[0063] Specifically, the mine microseismic monitoring data x is preprocessed by short-time Fourier transform (STFT) with preset number of items parameter k and preset window length parameter j to obtain the initial time-frequency domain features (usually symmetrical, including positive and negative frequencies). The positive frequency components in the initial time-frequency domain features are taken as the time-frequency domain features on the frequency axis. Represented as a four-dimensional tensor Where k is the number of frequency components and l / j / 2 is the number of time windows (or time resolution).

[0064] Furthermore, based on the time-frequency domain features Using t-layer multi-scale convolutional network (Multi Scale Convolution Networks, MSCN) A t Extract multi-scale time-frequency domain information of time-frequency domain features. Each layer Extract specific time domain scale 2 i-1 The characteristics of, where c is the number of hidden layers (or channels) of the network, 2 i-1 Indicates the change of time domain scale. As the number of layers increases, the time domain scale increases exponentially. The final output multi-scale time-frequency domain information is the concatenated tensor of all layer features in the last dimension, where represents the concatenation operation, which concatenates the features of all layers along the last dimension; The time domain scale is 2 i-1 The time-frequency domain features of the i-th convolutional subnetwork; c1 is the “total number of channels” or “total number of feature maps” of the network layer after concatenation, which may be equal to c×t (if the number of convolution kernels in each layer is the same and there is no additional processing), but may also be different due to the concatenation operation and possible dimensionality reduction or dimensionality increase operation; (2 t -1) represents the number of combinations of all time domain scales. Since the time domain scale grows exponentially, this number will increase rapidly with the increase of the number of layers; The time domain scale is 2 i-1 The time-frequency domain features of the i-th convolutional subnetwork.

[0065] The above method can effectively convert the mine microseismic monitoring data from the time domain to the time-frequency domain through short-time Fourier transform, thereby extracting the time-frequency domain features and using only the positive frequency components in the initial time-frequency domain features as the time-frequency domain features, which not only reduces the redundancy of the data but also improves the accuracy of feature extraction. The multi-scale time-frequency domain information extracted by the multi-size convolutional network can capture the characteristics of the data at different time scales, which helps to better understand the dynamic behavior and patterns of the data and improve the accuracy and reliability of data analysis.

[0066] Step S120, using multi-scale horizontal pooling to extract frequency domain energy distribution characteristics of mine microseismic monitoring data in the frequency domain.

[0067] In an optional embodiment, if Figure 3 As shown, step S120 includes:

[0068] Step S121, performing multi-scale horizontal pooling on the multi-scale time-frequency domain information in the frequency dimension to obtain a plurality of down-sampled feature maps;

[0069] Step S122, splicing each of the down-sampled feature maps along the frequency dimension to obtain the frequency domain energy distribution characteristics of the mine microseismic monitoring data.

[0070] Understandably, for multi-scale time-frequency information Each scale i (from 1 to r) of is divided into two i-1 The factor of downsampling is , that is, for scale i, the frequency dimension k will be divided into 2 i-1 parts, and perform multi-scale horizontal pooling on each part to obtain a downsampled feature map. Since there are r scales, r downsampled feature maps will be obtained, and the frequency dimension size of each downsampled feature map will be the original dimension.

[0071] Furthermore, each downsampled feature map is spliced ​​along the final frequency dimension to extract the frequency domain energy distribution characteristics of the mine microseismic monitoring data. in, is the time-frequency domain information In the frequency dimension k, press 2 i-1 Scale pooling result, c2 is the number of hidden layers in the network, (2 r -1) is the number of layers after multi-scale pooling.

[0072] The above method can effectively reduce the dimension of data by performing multi-scale horizontal pooling in the frequency dimension, thereby reducing the computational workload and storage requirements for subsequent processing. This dimensionality reduction operation not only simplifies the model but also improves computational efficiency. The frequency domain energy distribution characteristics of the extracted mine microseismic monitoring data provide richer and more accurate information for subsequent classification tasks.

[0073] Step S130, establishing a self-attention sequence model in the time domain, and extracting the time domain sequence features of the mine microseismic monitoring data according to the self-attention sequence model.

[0074] In an optional embodiment, if Figure 4 As shown, step S130 includes:

[0075] Step S131, pooling the self-attention sequence model and the frequency domain energy distribution feature to obtain a self-attention sequence model pooling result and a frequency domain energy distribution feature pooling result of the self-attention sequence model and the frequency domain energy distribution feature in the final frequency dimension;

[0076] Step S132, concatenating the pooling result of the self-attention sequence model with the pooling result of the frequency domain energy distribution feature to obtain the time domain sequence characteristics of the mine microseismic monitoring data.

[0077] Preferably, for multi-scale time-frequency information Build a sequence model with self-attention mechanism in its time domain This model may be a recurrent neural network (RNN), a long short-term memory network (LSTM), a gated recurrent unit (GRU), or other types of sequence processing models, which aims to extract time-domain sequence features at multiple scales in the time domain.

[0078] Furthermore, for the self-attention sequence model Energy distribution characteristics in frequency domain Pooling is performed to obtain a self-attention sequence model Energy distribution characteristics in frequency domain Pooling results of the self-attention sequence model on the last frequency dimension The pooling result of frequency domain energy distribution feature The above pooling operation may be global average pooling, global maximum pooling or other types of pooling operations, which aims to aggregate the features of each sensor and each scale into a single value or vector. Then the self-attention sequence model pooling result is concatenated The pooling result of frequency domain energy distribution feature Aggregate together to obtain the final time domain series features Expressed as

[0079] The above method achieves feature fusion by pooling and splicing the time domain sequence features extracted by the self-attention sequence model with the frequency domain energy distribution features. This fusion not only retains the important information of the time domain and frequency domain, but also removes redundant information through pooling operations, thereby enhancing the expressiveness of the features. In addition, by dynamically extracting features through the self-attention sequence model, there is no need to manually adjust the trigger parameters, and it can automatically adapt to the microseismic / blasting data classification requirements under different working conditions. There is no need to determine the time difference and time-frequency domain correlation through the channel arrival time, so as to avoid the influence of the arrival time picking accuracy in the previous process on the data classification accuracy.

[0080] Step S140, calculating the multi-sensor aggregation features in the mine microseismic monitoring data according to the frequency domain energy distribution features and the time domain sequence features.

[0081] In an optional embodiment, if Figure 5 As shown, step S140 includes:

[0082] Step S141, performing multi-sensor feature extraction on the frequency domain energy distribution feature and the time domain sequence feature to obtain a feature vector;

[0083] Step S142, pooling the feature vector in the sensor dimension to obtain the multi-sensor aggregate feature.

[0084] Specifically, the frequency domain energy distribution characteristics and time domain sequence characteristics of each sensor are extracted to obtain a new feature vector Its dimensions are still n×m×(2×c2). Here represents the time domain sequence characteristics and frequency domain energy distribution characteristics of the original mine microseismic monitoring data, and S represents the feature extraction operation, which may include filtering, transformation (such as Fourier transform), feature selection or dimensionality reduction steps, which will not be described in detail in this embodiment.

[0085] Furthermore, the extracted feature vector Pooling is performed on the sensor dimension (i.e., m dimension) to aggregate information from different sensors. Pooling operation P n×m×(2×c2)→n×(2×c2) Aggregate m-dimensional data into a single value or vector to obtain multi-sensor aggregate features The above process can be expressed as:

[0086] It should be noted that the above pooling operation can be maximum pooling (taking the maximum value in each window), average pooling (taking the average value in each window) or other types of pooling methods, which is not limited in this embodiment.

[0087] The above method can more comprehensively capture the information in the mine microseismic monitoring data by fusing the frequency domain energy distribution features and the time domain sequence features, improve the feature expression ability, and effectively integrate the information from different sensors. This helps to eliminate the redundancy and differences between sensors, improve the overall consistency and reliability of the data, and enhance the performance of the model in processing the task of adaptive automatic classification of mine microseismic monitoring data.

[0088] Step S150, using a multi-layer perceptron to classify the multi-sensor aggregate features, and using a feature matching method to perform positive and negative sample constraints on the multi-sensor aggregate features, to obtain a classification result of the mine microseismic monitoring data.

[0089] In an optional embodiment, if Figure 6 As shown, step S150 includes:

[0090] Step S151, preprocessing and positive and negative sample constraints are performed on the multi-sensor aggregated features to obtain optimized multi-sensor aggregated features, and inputting the optimized multi-sensor aggregated features into the multi-layer perceptron for classification;

[0091] Step S152: converting the output result of the multilayer perceptron into a probability distribution using an activation function, and obtaining the classification result according to the probability distribution.

[0092] Specifically, for multi-sensor aggregation features Appropriate preprocessing (such as standardization, normalization, etc.) is performed to ensure that the data is suitable for input into the Multi-Layer Perceptron (MLP). To standardize, Among them, μ and σ are the multi-sensor aggregation features, The mean and standard deviation of .

[0093] Then, if Figure 7 As shown in the figure, the feature matching method is used to aggregate the preprocessed multi-sensor features. Perform positive and negative sample constraints (positive and negative samples can be defined based on known target features and background features) to obtain optimized multi-sensor aggregate features, and then input the optimized multi-sensor aggregate features into MLP. MLP is usually composed of multiple fully connected layers, each of which contains a weight matrix W and a bias vector b. Assuming that the output of MLP is y, then: Here, L represents the number of layers of MLP and θ represents the activation function (such as ReLU).

[0094] Furthermore, an activation function (such as Softmax) is used in the output layer of the MLP to convert the output result y into a probability distribution P, and then the category with the highest probability is selected as the final classification result, that is, Among them, c represents the number of types in the output results, p i represents the probability of the i-th category.

[0095] The above method accurately classifies the mine microseismic monitoring data through MLP. Since MLP has multiple hidden layers, it can capture the high-order features in the data, which enhances the robustness of the model to noise and outliers, making the classification results more stable and reliable. Using the activation function in the output layer, MLP can convert the output results into probability distribution, which not only provides confidence information for the classification results, but also helps to make more informed decisions in multi-classification tasks.

[0096] In an alternative embodiment, Figure 8 As shown, the method for adaptive automatic classification of mine microseismic monitoring data in this embodiment also includes:

[0097] Step S160, obtaining a type label vector of the mine microseismic monitoring data, and performing short-time Fourier transform on the type label vector to obtain a label vector;

[0098] Step S170, calculating the cross entropy loss between the classification result and the label vector, and using the cross entropy loss as the feature classification loss;

[0099] Step S180, obtaining data of the same category and data of different categories of the mine microseismic monitoring data, calculating the triplet loss of the mine microseismic monitoring data, the data of the same category and the data of different categories based on the multi-scale time-frequency domain information, the frequency domain energy distribution characteristics, the time domain sequence characteristics and the multi-sensor aggregation characteristics, and using the triplet loss as the feature matching loss;

[0100] Step S190: jointly constraining the adaptive automatic classification model of mine microseismic monitoring data according to the feature classification loss and the feature matching loss.

[0101] Understandably, after outputting the classification results of the mine microseismic monitoring data, the model converts the classification results into one-hot encoding form and uses it as the category label vector gt(x) of the mine microseismic monitoring data, that is, a vector of length c, in which only one element is 1 (indicating the classification result) and the rest are 0. For example, if there are three categories of data output, namely microseismic data, blasting data, and noise data, then their corresponding one-hot encodings are microseismic [1, 0, 0], blasting [0, 1, 0], and noise [0, 0, 1]. Further, the category label vector gt(x) is short-time Fourier transformed to obtain the label vector

[0102] According to the features of each layer obtained in the above steps S110 to S140 It contains scale time-frequency domain information Frequency domain energy distribution characteristics Time domain series characteristics Aggregate features with multiple sensors Right now And the classification results Aggregate into a data set D. Rationally and evenly divide the data set D into training set D train , validation set D valid , test set D test .

[0103] For the training set D train For each microseismic monitoring data x in (whose category label vector is gt(x)), calculate the classification result and label vector The cross entropy loss between The cross entropy loss is used as the feature classification loss, which measures the difference between the probability distribution predicted by the model and the true label distribution. The smaller the cross entropy loss, the more accurate the model's prediction.

[0104] For the training set D train For each microseismic monitoring data x in the + and heterogeneous data x - , construct a triple data based on Calculating triplet loss The triplet loss is used as the feature matching loss, which measures the relative size of the distance between the same category data and the distance between different categories of data. Dist represents the distance between the feature tensors of each layer (such as Euclidean distance or cosine distance), and m is the relaxation coefficient for adjusting the constraint ability. fp (x t) is used to measure the relative size of the distance between data of the same category and the distance between data of different categories. Specifically, it hopes that the distance between data of the same category As small as possible, and the distance between different categories of data As big as possible.

[0105] In order to take into account the time-frequency domain representation ability of mine microseismic monitoring data and the ability to distinguish different types of data, the training loss L train Using feature classification loss L cls And feature matching loss L fp The weighted sum of train =L cls +αL fp , where α is the weight of the feature matching loss, which is used to balance the contribution of the two loss terms.

[0106] The training loss achieves a joint constraint on the model performance by combining feature classification loss and feature matching loss. Feature classification loss ensures that the model can accurately classify the input data, while feature matching loss enhances the model representation's ability to distinguish between different categories of data, avoids overfitting on known training data, and retains a certain recognition ability for unknown data.

[0107] It should be noted that the validation set D valid During the training process, it can be used to adjust the model's hyperparameters (such as learning rate, weight, etc.). The validation set does not participate in the model training, but is used to evaluate the model's performance during training. After the model training is completed, the test set D test The test set is used to evaluate the classification accuracy of the model. It also does not participate in the training of the model and will not be described in detail in this embodiment.

[0108] Traditional adaptive automatic classification methods for mine microseismic monitoring data are based on limited types of data training, and only consider the model's feature extraction and classification capabilities on existing data sets. They are prone to losing the ability to represent mine microseismic monitoring data due to overfitting, and thus cannot cope with unknown data from complex vibration sources in actual working environments. The above method improves the model training process by combining feature classification loss and feature matching loss, retains the classification method's ability to represent mine microseismic monitoring data, and improves the model's classification performance for mine microseismic monitoring data.

[0109] The method for adaptive automatic classification of mine microseismic monitoring data provided in the embodiment of the present application obtains mine microseismic monitoring data, performs short-time Fourier transform on the mine microseismic monitoring data, obtains time-frequency domain features, and uses a time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain features; uses multi-scale horizontal pooling to extract frequency domain energy distribution features of mine microseismic monitoring data in the frequency domain; establishes a self-attention sequence model in the time domain, and extracts time domain sequence features of the mine microseismic monitoring data according to the self-attention sequence model; calculates multi-sensor aggregation features in the mine microseismic monitoring data according to the frequency domain energy distribution features and the time domain sequence features; uses a multi-layer perceptron to classify the multi-sensor aggregation features, and uses a feature matching method to constrain positive and negative samples on the multi-sensor aggregation features to obtain the classification results of the mine microseismic monitoring data. The present application can improve the timeliness and accuracy of data processing, realize effective extraction and fusion of features, and improve the accuracy and reliability of classification results.

[0110] Example 2

[0111] like Fig. 9 FIG. 1 is a schematic diagram of a structure of a device 900 for adaptively automatically classifying microseismic monitoring data of a mine in an embodiment of the present application, wherein the device comprises:

[0112] A shallow feature extraction module 910 is used to obtain microseismic monitoring data of a mine, perform short-time Fourier transform on the microseismic monitoring data of the mine to obtain time-frequency domain features, and use a time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain features;

[0113] A frequency domain energy distribution feature extraction module 920 is used to extract the frequency domain energy distribution features of the mine microseismic monitoring data in the frequency domain by using multi-scale horizontal pooling;

[0114] A time domain sequence feature extraction module 930 is used to establish a self-attention sequence model in the time domain, and extract the time domain sequence features of the mine microseismic monitoring data according to the self-attention sequence model;

[0115] A multi-channel aggregation feature calculation module 940 is used to calculate the multi-sensor aggregation features in the mine microseismic monitoring data according to the frequency domain energy distribution features and the time domain sequence features;

[0116] The classification module 950 is used to classify the multi-sensor aggregate features using a multi-layer perceptron, and use a feature matching method to perform positive and negative sample constraints on the multi-sensor aggregate features to obtain the classification results of the mine microseismic monitoring data.

[0117] Optionally, the above-mentioned mine microseismic monitoring data adaptive automatic classification device 900 further includes:

[0118] A first transformation module is used to obtain a preset number of items parameter and a preset window length parameter, and perform a short-time Fourier transform on the mine microseismic monitoring data according to the preset number of items parameter and the preset window length parameter to obtain initial time-frequency domain features;

[0119] The first extraction module is used to use the positive frequency component in the initial time-frequency domain feature as the time-frequency domain feature, and use the time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain feature.

[0120] Optionally, the above-mentioned mine microseismic monitoring data adaptive automatic classification device 900 further includes:

[0121] A first pooling module, configured to perform multi-scale horizontal pooling on the multi-scale time-frequency domain information in the frequency dimension to obtain a plurality of down-sampled feature maps;

[0122] The first splicing module is used to splice each of the down-sampled feature maps along the frequency dimension to obtain the frequency domain energy distribution characteristics of the mine microseismic monitoring data.

[0123] Optionally, the above-mentioned mine microseismic monitoring data adaptive automatic classification device 900 further includes:

[0124] A second pooling module is used to pool the self-attention sequence model and the frequency domain energy distribution feature to obtain a self-attention sequence model pooling result and a frequency domain energy distribution feature pooling result of the self-attention sequence model and the frequency domain energy distribution feature in the final frequency dimension;

[0125] The second splicing module is used to splice the pooling result of the self-attention sequence model with the pooling result of the frequency domain energy distribution feature to obtain the time domain sequence characteristics of the mine microseismic monitoring data.

[0126] Optionally, the above-mentioned mine microseismic monitoring data adaptive automatic classification device 900 further includes:

[0127] A second extraction module is used to perform multi-sensor feature extraction on the frequency domain energy distribution feature and the time domain sequence feature to obtain a feature vector;

[0128] The third pooling module is used to pool the feature vector in the sensor dimension to obtain the multi-sensor aggregate feature.

[0129] Optionally, the above-mentioned mine microseismic monitoring data adaptive automatic classification device 900 further includes:

[0130] A preprocessing module is used to perform preprocessing and positive and negative sample constraints on the multi-sensor aggregate features to obtain optimized multi-sensor aggregate features, and input the optimized multi-sensor aggregate features into the multi-layer perceptron for classification;

[0131] The conversion module is used to convert the output result of the multilayer perceptron into a probability distribution by using an activation function, and obtain the classification result according to the probability distribution.

[0132] Optionally, the above-mentioned mine microseismic monitoring data adaptive automatic classification device 900 further includes:

[0133] A second transformation module is used to obtain a type label vector of the mine microseismic monitoring data, and perform a short-time Fourier transform on the type label vector to obtain a label vector;

[0134] A first calculation module, used to calculate the cross entropy loss between the classification result and the label vector, and use the cross entropy loss as the feature classification loss;

[0135] A second calculation module is used to obtain the same category data and different category data of the mine microseismic monitoring data, calculate the triplet loss of the mine microseismic monitoring data, the same category data and the different category data based on the multi-scale time-frequency domain information, the frequency domain energy distribution characteristics, the time domain sequence characteristics and the multi-sensor aggregation characteristics, and use the triplet loss as the feature matching loss;

[0136] The constraint module is used to jointly constrain the adaptive automatic classification model of mine microseismic monitoring data according to the feature classification loss and the feature matching loss.

[0137] The adaptive automatic classification device for mine microseismic monitoring data provided in the embodiment of the present application can improve the timeliness and accuracy of data processing, realize the effective extraction and fusion of features, and improve the accuracy and reliability of classification results.

[0138] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and structure diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in an alternative implementation, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the structure diagram and / or the flow chart, and the combination of boxes in the structure diagram and / or the flow chart, can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0139] In addition, the functional modules or units in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0140] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or partly contributed to the prior art or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a smart phone, a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium can be a non-volatile storage medium or a volatile storage medium. For example, the storage medium can be: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, and other media that can store program codes.

[0141] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for adaptive automatic classification of mine microseismic monitoring data, characterized in that: Applied to the adaptive automatic classification model of mine microseismic monitoring data, the method includes: Acquire microseismic monitoring data of a mine, perform short-time Fourier transform on the microseismic monitoring data of the mine to obtain time-frequency domain features, and use a time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain features; Multi-scale horizontal pooling is used to extract frequency domain energy distribution characteristics of the mine microseismic monitoring data in the frequency domain; Establishing a self-attention sequence model in the time domain, and extracting the time domain sequence features of the mine microseismic monitoring data according to the self-attention sequence model; Calculating multi-sensor aggregation features in the mine microseismic monitoring data according to the frequency domain energy distribution features and the time domain sequence features; A multi-layer perceptron is used to classify the multi-sensor aggregate features, and a feature matching method is used to constrain the multi-sensor aggregate features by positive and negative samples, so as to obtain the classification results of the mine microseismic monitoring data; The method of extracting the frequency domain energy distribution characteristics of the mine microseismic monitoring data in the frequency domain by using multi-scale horizontal pooling includes: Performing multi-scale horizontal pooling on the multi-scale time-frequency domain information in the frequency dimension to obtain multiple down-sampled feature maps; The down-sampled feature graphs are spliced ​​along the frequency dimension to obtain the frequency domain energy distribution characteristics of the mine microseismic monitoring data; The step of extracting the time domain sequence features of the mine microseismic monitoring data according to the self-attention sequence model includes: Pooling the self-attention sequence model and the frequency domain energy distribution feature to obtain a self-attention sequence model pooling result and a frequency domain energy distribution feature pooling result of the self-attention sequence model and the frequency domain energy distribution feature in the final frequency dimension; The pooling result of the self-attention sequence model is concatenated with the pooling result of the frequency domain energy distribution feature to obtain the time domain sequence feature of the mine microseismic monitoring data; The method for adaptive automatic classification of mine microseismic monitoring data also includes: Obtaining a type label vector of the mine microseismic monitoring data, and performing short-time Fourier transform on the type label vector to obtain a label vector; Calculating the cross entropy loss between the classification result and the label vector, and using the cross entropy loss as the feature classification loss; Acquire the same category data and different category data of the mine microseismic monitoring data, calculate the triplet loss of the mine microseismic monitoring data, the same category data and the different category data based on the multi-scale time-frequency domain information, the frequency domain energy distribution characteristics, the time domain sequence characteristics and the multi-sensor aggregation characteristics, and use the triplet loss as the feature matching loss; The adaptive automatic classification model of mine microseismic monitoring data is jointly constrained according to the feature classification loss and the feature matching loss.

2. The method for adaptive automatic classification of mine microseismic monitoring data according to claim 1 is characterized in that: The short-time Fourier transform is performed on the mine microseismic monitoring data to obtain time-frequency domain features, and a time-domain multi-scale convolutional network is used to extract multi-scale time-frequency domain information of the time-frequency domain features, including: Obtaining a preset number of item parameters and a preset window length parameter, and performing a short-time Fourier transform on the mine microseismic monitoring data according to the preset number of item parameters and the preset window length parameter to obtain initial time-frequency domain features; The positive frequency component in the initial time-frequency domain feature is used as the time-frequency domain feature, and the time-domain multi-scale convolutional network is used to extract multi-scale time-frequency domain information of the time-frequency domain feature.

3. The method for adaptive automatic classification of mine microseismic monitoring data according to claim 1, characterized in that: The calculating, according to the frequency domain energy distribution characteristics and the time domain sequence characteristics, the multi-sensor aggregation characteristics in the mine microseismic monitoring data comprises: Performing multi-sensor feature extraction on the frequency domain energy distribution feature and the time domain sequence feature to obtain a feature vector; The feature vector is pooled in the sensor dimension to obtain the multi-sensor aggregate feature.

4. The method for adaptive automatic classification of mine microseismic monitoring data according to claim 1, characterized in that: The multi-layer perceptron is used to classify the multi-sensor aggregate features, and a feature matching method is used to constrain the multi-sensor aggregate features by positive and negative samples to obtain the classification results of the mine microseismic monitoring data, including: Preprocessing and constraining the multi-sensor aggregate features with positive and negative samples to obtain optimized multi-sensor aggregate features, and inputting the optimized multi-sensor aggregate features into the multi-layer perceptron for classification; The output result of the multilayer perceptron is converted into a probability distribution using an activation function, and the classification result is obtained according to the probability distribution.

5. An adaptive automatic classification device for mine microseismic monitoring data, characterized in that: Applied to the adaptive automatic classification model of mine microseismic monitoring data, the device comprises: A shallow feature extraction module is used to obtain mine microseismic monitoring data, perform short-time Fourier transform on the mine microseismic monitoring data to obtain time-frequency domain features, and use a time-domain multi-scale convolutional network to extract multi-scale time-frequency domain information of the time-frequency domain features; A frequency domain energy distribution feature extraction module, used to extract the frequency domain energy distribution features of the mine microseismic monitoring data in the frequency domain by using multi-scale horizontal pooling; A time domain sequence feature extraction module is used to establish a self-attention sequence model in the time domain, and to extract the time domain sequence features of the mine microseismic monitoring data according to the self-attention sequence model; A multi-channel aggregation feature calculation module, used to calculate the multi-sensor aggregation features in the mine microseismic monitoring data according to the frequency domain energy distribution features and the time domain sequence features; A classification module is used to classify the multi-sensor aggregate features using a multi-layer perceptron, and to perform positive and negative sample constraints on the multi-sensor aggregate features using a feature matching method, so as to obtain a classification result of the mine microseismic monitoring data; Wherein, the frequency domain energy distribution feature extraction module also includes: A first pooling module, configured to perform multi-scale horizontal pooling on the multi-scale time-frequency domain information in the frequency dimension to obtain a plurality of down-sampled feature maps; A first splicing module is used to splice each of the down-sampled feature maps along the frequency dimension to obtain the frequency domain energy distribution characteristics of the mine microseismic monitoring data; Wherein, the time domain sequence feature extraction module also includes: A second pooling module is used to pool the self-attention sequence model and the frequency domain energy distribution feature to obtain a self-attention sequence model pooling result and a frequency domain energy distribution feature pooling result of the self-attention sequence model and the frequency domain energy distribution feature in the final frequency dimension; A second splicing module is used to splice the pooling result of the self-attention sequence model with the pooling result of the frequency domain energy distribution feature to obtain the time domain sequence characteristics of the mine microseismic monitoring data; The self-adaptive automatic classification device for mine microseismic monitoring data also includes: A second transformation module is used to obtain a type label vector of the mine microseismic monitoring data, and perform a short-time Fourier transform on the type label vector to obtain a label vector; A first calculation module, used to calculate the cross entropy loss between the classification result and the label vector, and use the cross entropy loss as the feature classification loss; A second calculation module is used to obtain the same category data and different category data of the mine microseismic monitoring data, calculate the triplet loss of the mine microseismic monitoring data, the same category data and the different category data based on the multi-scale time-frequency domain information, the frequency domain energy distribution characteristics, the time domain sequence characteristics and the multi-sensor aggregation characteristics, and use the triplet loss as the feature matching loss; The constraint module is used to jointly constrain the adaptive automatic classification model of mine microseismic monitoring data according to the feature classification loss and the feature matching loss.

Images