A small sample-oriented microseismic signal classification method fusing diffusion enhancement and domain adaptation

Abstract

The embodiment of the application provides a small sample-oriented microseismic signal classification method combining diffusion enhancement and domain self-adaptation, relates to the technical field of microseismic signal processing, combines diffusion enhancement and domain self-adaptation mechanism in a microseismic signal classification model, increases the number of samples and enriches sample characteristics by amplifying coal mine microseismic event and blasting event training samples, alleviates the limitation of sample scarcity on model training, and reduces the overfitting risk under small sample conditions; through the domain self-adaptation mechanism, the influence of signal distribution difference caused by different working conditions and monitoring equipment state changes on the classification result is reduced, and the adaptability of the model to different monitoring conditions is enhanced; through the adjustment and update of model parameters in the training process, the distinguishing ability and generalization performance of the model to microseismic events and blasting events are improved, so that the accuracy and stability of microseismic signal classification under small sample conditions are improved, and signal classification basis is provided for mine seismic source positioning, energy calculation and disaster risk discrimination.

Description

Technical Field

[0001] This application relates to the field of microseismic signal processing technology, specifically to a microseismic signal classification method based on fusion diffusion enhancement and domain adaptation for small samples. Background Technology

[0002] With the rapid development of intelligent mining and smart geological monitoring technologies, safe production has become a core issue in underground engineering and energy extraction. Microseismic monitoring, as a highly sensitive early warning method for dynamic disasters, can capture the energy release process within rock masses in real time, providing crucial data for rockburst prediction, coal seam gas outburst identification, and geological tectonic activity analysis. Signal identification, as the primary and key technology in microseismic monitoring data processing, directly determines the reliability of subsequent source location, energy calculation, and risk assessment.

[0003] However, the mining and underground engineering environment has its own unique complexities: on the one hand, continuous engineering activities such as mining and tunneling generate a large number of blasting and mechanical vibration signals that are similar to the characteristics of real disaster signals. On the other hand, strong environmental noise can severely pollute or even drown out effective signals, causing distortion of their waveform characteristics. Crucially, microseismic events and blasting events, which are precursory signals of disasters with clear physical significance, are rare events in actual monitoring data, making it difficult to collect and label them in large quantities. This results in severe sample scarcity and class imbalance in the dataset. Consequently, traditional identification methods relying on fixed thresholds generally suffer from insufficient adaptability and poor stability, directly affecting the accuracy and timeliness of disaster early warning.

[0004] The development of deep learning has provided new ideas for the automatic identification and classification of microseismic signals. For example, patent application CN116990865A discloses a microseismic event detection method and system based on deep transfer learning. Patent application CN113837116A discloses a transfer learning method and system for microseismic signal identification models. However, traditional neural network models with fixed structures have limited generalization ability on small sample training sets, and lack the ability to adaptively model the dynamic characteristics and temporal correlation of signals when dynamic working conditions or equipment status changes, resulting in a significant decrease in classification performance. Summary of the Invention

[0005] In view of this, this application provides a microseismic signal classification method based on fusion diffusion enhancement and domain adaptation for small samples, in order to solve the problem of poor microseismic signal classification performance in the prior art.

[0006] In a first aspect, embodiments of this application provide a microseismic signal classification method for small-sample fusion diffusion enhancement and domain adaptation, including: The acquired raw microseismic monitoring data of the coal mine was preprocessed, and the microseismic event signals and blasting event signals were extracted as training samples. Construct a microseismic signal classification model that integrates diffusion enhancement and domain adaptation mechanisms; After amplifying the training samples, the microseismic signal classification model is trained. During training, the model parameters are adjusted and updated to make the model converge and obtain the optimal model weight parameters. Microseismic signals are classified using the microseismic signal classification model updated with the optimal model weight parameters.

[0007] In one possible implementation, the preprocessing of the acquired raw microseismic monitoring data from the coal mine, extracting microseismic event signals and blasting event signals as training samples, includes: The raw microseismic monitoring data of the target monitoring area is acquired and preprocessed, including denoising, normalization and signal extraction, to construct a raw sample set containing microseismic signals and their corresponding category labels. Microseismic event signals and blasting event signals were extracted from the original sample set as training samples.

[0008] In one possible implementation, the microseismic signal classification model includes: a diffusion enhancement network, a temporal feature extraction network, a domain adaptive discriminant network, and a microseismic signal classification network. The construction of the microseismic signal classification model integrating diffusion enhancement and domain adaptive mechanisms includes: A forward diffusion process is established to gradually inject Gaussian noise into the original signal, and a denoising network conditioned on category labels and time steps is constructed. The forward diffusion process is then combined to form a diffusion enhancement network. Source domain data is constructed based on the signal samples generated by the diffusion enhancement network and the original signal samples, and target domain data is obtained by adding Gaussian white noise to the source domain data. A temporal feature extraction network is constructed using multi-layer stacked bidirectional gated recurrent units. The source domain data and target domain data are connected to the same temporal feature extraction network, and the concatenation result of the forward and backward hidden states is used as the feature output. Construct a domain adaptive discriminant network that includes a gradient reversal layer and a domain discriminant network, and connect it to the output of the temporal feature extraction network to establish a domain adversarial training branch; A multi-layer fully connected network is used to construct a microseismic signal classification network, which is then connected to the output of a temporal feature extraction network. The output dimension of this network is set to the number of signal categories.

[0009] In one possible implementation, the step of constructing source domain data based on the signal samples generated by the diffusion enhancement network and the original signal samples, and obtaining target domain data by adding Gaussian white noise to the source domain data, includes: Conditional sampling is performed using a trained denoising network to sample initial noise from a standard Gaussian distribution. By specifying target category conditions and performing iterative denoising using reverse diffusion, realistic synthetic microseismic and blast signals are generated. in: This is the signal after denoising in the previous step. The current noise signal, For time steps, For target category conditions, The cumulative coefficient is... To predict noise components, The noise standard deviation of the reverse process. Standard Gaussian noise; The generated samples are injected into the original dataset to construct a class-balanced augmented dataset as the source domain data; Additive white Gaussian noise is injected into the source domain data to form the target domain data, which is used to simulate noisy working conditions in actual monitoring.

[0010] In one possible implementation, the temporal feature extraction network is constructed using multi-layered stacked bidirectional gated recurrent units. The source domain data and target domain data are connected to the same temporal feature extraction network, and the concatenation result of the forward and backward hidden states is used as the feature output, including: A multi-layer bidirectional gated loop unit is constructed for timing modeling, and source domain data and target domain data are input into the multi-layer bidirectional gated loop unit; The bidirectional gated loop unit is configured as a multi-layer stacked structure, at each time step The bidirectional gated loop unit calculates the forward hidden state respectively. and backward hidden state ; After processing by the bidirectional gated recurrent unit, the output of the last time step is extracted as a high-dimensional temporal feature vector. The calculation formula is: .

[0011] In one possible implementation, the construction of a domain adaptive discriminant network comprising a gradient reversal layer and a domain discriminant network, and connecting it to the output of a temporal feature extraction network to establish a domain adversarial training branch, includes: The First, the gradient inversion layer (GRL) processes the feature extractor, enabling it to learn common features that are insensitive to the domain. The features processed by GRL are input into a domain adaptive discriminant network, which consists of: a fully connected layer, a LeakyReLU activation function, a Dropout layer, a fully connected layer, a LeakyReLU activation function, and a final domain classification fully connected layer. During adversarial training, the parameters of the domain adaptive discriminant network are updated in the direction that reduces the domain discrimination error, while the parameters of the feature extractor are updated in the direction that increases the domain discrimination error under the action of the gradient reversal layer. Ultimately, the feature extractor converges to domain-invariant features that are insensitive to domain information and only retain information related to class discrimination. .

[0012] In one possible implementation, after amplifying the training samples, the microseismic signal classification model is trained. During training, the model parameters are adjusted and updated to bring the model to converge and obtain the optimal model weight parameters, including: The training samples were amplified and then divided into a training set, a validation set, and a test set. The microseismic signal classification model is trained using the training set. The model parameters are adjusted using the validation set. The model parameters are updated using the backpropagation algorithm to bring the model to converge and obtain the optimal model weight parameters.

[0013] One possible implementation also includes: Construct a composite loss function for optimizing and updating model parameters, with the classification loss employing the multi-class cross-entropy loss function. The domain discriminator uses a binary cross-entropy loss function to optimize the domain adaptation loss. The noise prediction loss of the diffusion model is calculated using the mean square error loss. ; During the training phase of the diffusion-enhanced network, with Optimize the parameters of the denoising network; during the training phase of the classification model, jointly optimize the feature extractor, classifier, and domain adaptive discriminant network using the total loss function; The model training parameters are set, and the AdamW optimization algorithm is used to optimize the model parameter learning. During the training process, the learning rate is gradually reduced, and after iterative training to the preset number of training rounds, the optimal model weights are saved.

[0014] In one possible implementation, the total loss function is expressed as: Multi-class cross-entropy loss function for: The binary cross-entropy loss function is used to optimize the domain adaptation loss. for: Noise prediction loss for: in: The weighted hyperparameters for controlling the alignment strength of the control domain, For batch size, For the number of categories, For the first The sample at the th The true label of the class, The model predicts the first The sample belongs to the first The probability of a class; For real labels, The domain probability predicted by the model; The noise predicted by the diffusion model, This is real noise.

[0015] In one possible implementation, the microseismic signal classification using the microseismic signal classification model updated with the optimal model weight parameters includes: The test set is input into a microseismic signal classification model loaded with the optimal model weights for testing; During the testing phase, the domain adaptive discrimination module was completely discarded, and only the bidirectional gated recurrent unit feature extractor and multi-classifier were retained to perform forward propagation. Noise-added test data were constructed according to the preset signal-to-noise ratio levels to evaluate the classification performance of the model under different noise intensities.

[0016] Secondly, embodiments of this application provide a microseismic signal classification system for small sample fusion diffusion enhancement and domain adaptation, including: The data processing module is used to preprocess the acquired raw microseismic monitoring data of the coal mine and extract the microseismic event signals and blasting event signals as training samples. The model building module is used to construct a microseismic signal classification model that integrates diffusion enhancement and domain adaptation mechanisms; The weight parameter determination module is used to amplify the training samples and then train the microseismic signal classification model. During training, the model parameters are adjusted and updated to make the model converge and obtain the optimal model weight parameters. The microseismic signal classification network is used to classify microseismic signals using a microseismic signal classification model updated with the optimal model weight parameters.

[0017] Thirdly, embodiments of this application provide an electronic device, including: processor; Memory; And a computer program, wherein the computer program is stored in the memory, the computer program including instructions that, when executed by the processor, cause the electronic device to perform the method described in any one of the first aspects.

[0018] Fourthly, embodiments of this application provide a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the method described in any one of the first aspects.

[0019] In this embodiment, diffusion enhancement and domain adaptation mechanisms are integrated into the microseismic signal classification model. By amplifying the training samples of coal mine microseismic events and blasting events, the number of samples is increased and the sample features are enriched, alleviating the limitation of model training caused by sample scarcity and reducing the risk of overfitting under small sample conditions. Through the domain adaptation mechanism, the impact of signal distribution differences caused by changes in different working conditions and monitoring equipment status on the classification results is reduced, enhancing the model's adaptability to different monitoring conditions. By adjusting and updating the model parameters during training, the model's ability to distinguish between microseismic events and blasting events and its generalization performance are improved, thereby enhancing the accuracy and stability of microseismic signal classification under small sample conditions and providing signal classification basis for mine seismic source location, energy calculation, and disaster risk assessment. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a microseismic signal classification method based on fusion diffusion enhancement and domain adaptation for small samples, provided in an embodiment of this application; Figure 2 A schematic diagram of a diffusion enhancement network is provided for embodiments of this application; Figure 3 A schematic diagram of the temporal feature extraction network, the domain adaptive discriminant network, and the microseismic signal classification network provided in the embodiments of this application; Figure 4 A schematic diagram of a microseismic signal classification system for small sample fusion diffusion enhancement and domain adaptation provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0022] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0023] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0024] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0025] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0026] Example 1 See Figure 1 This embodiment provides a microseismic signal classification method based on fusion diffusion enhancement and domain adaptation for small samples, including: S101, the acquired raw microseismic monitoring data of the coal mine is preprocessed, and the microseismic event signals and blasting event signals are extracted as training samples.

[0027] The microseismic signal classification dataset used in this method originates from real monitoring data collected in a coal mine in Shanxi Province. This dataset is specifically designed for microseismic event identification tasks under complex deep working conditions. It covers three typical signal categories commonly found in deep working conditions: microseismic events, blasting events, and environmental noise. To distinguish and manage different signal categories, all signal samples are pre-organized and stored in different category folders. The model's category label information is directly obtained from the folder name where the file is located, eliminating the need for additional label annotation files.

[0028] To ensure the standardization and consistency of the model input, all original signal samples underwent uniform preprocessing operations, including baseline correction, denoising, and uniform signal truncation, ensuring that each signal remained consistent in terms of time length and sampling frequency. This resulted in the construction of a standard temporal input tensor for subsequent deep learning classification tasks.

[0029] Using the microseismic event signals and blasting event signals selected from the processed sample set, the subsequently constructed diffusion enhancement network is iteratively trained so that the diffusion enhancement network learns the characteristic distribution of the microseismic event signals and blasting event signals.

[0030] S102, Construct a microseismic signal classification model that integrates diffusion enhancement and domain adaptation mechanisms.

[0031] A microseismic signal classification model integrating diffusion enhancement and domain adaptation mechanisms is constructed. The microseismic signal classification model mainly includes a diffusion enhancement network, a temporal feature extraction network, a domain adaptation discrimination network, and a microseismic signal classification network.

[0032] See Figure 2 In the diffusion-enhanced network, a forward Markov chain is constructed for any time step. The state transition equation is defined using the reparameterization technique: ,in The coefficients are cumulative, meaning the α coefficients from the first t time steps are multiplied together. This represents standard normally distributed noise. For the input real microseismic signal sample... Set the total number of diffusion steps Follow the preset variance The scheduling strategy gradually injects Gaussian noise. .

[0033] For the output microseismic signal data with injected Gaussian noise Match it with the corresponding time step Embedded vectors and category labels The embedding vectors are used as input to a one-dimensional temporal U-Net conditional denoising network. Input data Multi-level one-dimensional convolution and activation functions are employed to extract multi-scale time-frequency features, with the number of channels progressively expanded layer by layer. Max pooling is used to compress the temporal dimension of the signal, forming deep features for the encoder. A self-attention mechanism is introduced at the bottleneck layer of U-Net to model the long-distance temporal dependencies of deep features, enhancing the global coherence of the signal. The temporal resolution of the signal is gradually restored through deconvolution layers to achieve feature decoding. Skip connections are used to fuse shallow features from different levels of the encoder to preserve signal detail, ultimately outputting the predicted noise component. .

[0034] Using a pre-trained conditional denoising network Conditional sampling is performed to generate missing microseismic signal samples; firstly, samples are obtained from a standard Gaussian distribution. Mid-sample initial noise and specify the target category conditions. Used to guide the generation process; initial noise and specified category conditions Input into the trained denoising network In the middle. Following the reverse diffusion formula. The model is from Iterate to Noise reduction is performed to ultimately output a realistic, high-fidelity synthetic microseismic signal. The generated synthetic sample The samples are fused with those from the original training dataset; by adjusting the number of synthetic samples, an augmented dataset with balanced class distribution is constructed. This is used as input for training subsequent microseismic signal classification models.

[0035] Based on this, domain-adaptive training data is constructed: the aforementioned class-balanced augmented dataset is used as the source domain data, with its domain label denoted as 0. Additive white Gaussian noise is injected into the source domain signal to form the target domain data, used to simulate noisy working conditions in actual monitoring; its domain label is denoted as 1. The target domain signal is generated by the following formula: Where ε is Gaussian white noise with a mean of 0 and a variance of σ²; the noise variance σ² is determined by the preset signal-to-noise ratio (SNR) according to the following formula: in, SNR is the average power of the source domain signal, which is the mean of the squares of the amplitudes at each sampling point of the signal; SNR is the preset signal-to-noise ratio, in decibels (dB).

[0036] During training, the signal-to-noise ratio (SNR) of each sample is randomly sampled within the range of [0, 20] dB. During validation, noise is injected at six SNR levels (20 dB, 15 dB, 10 dB, 5 dB, 0 dB, -5 dB) to construct validation data for hyperparameter tuning and optimal model selection. During testing, noise is injected at six SNR levels (20 dB, 15 dB, 10 dB, 5 dB, 0 dB, -5 dB) to construct test data for evaluation. The source domain data and target domain data together constitute the domain adaptive dataset. .

[0037] See Figure 3 For domain-adaptive datasets The input is fed into a temporal feature extraction network constructed using multi-layered stacked bidirectional gated recurrent units (BRNs) for temporal modeling. Source and target domain samples share the same multi-layered BRNs for feature encoding to ensure subsequent domain alignment is performed in a unified feature space. The BRNs are configured as a 3-layer stacked structure with hidden layer dimensions... It is configured to be bidirectional to capture the context of the signal.

[0038] At each time step The bidirectional gated loop unit calculates the forward hidden state respectively. and backward hidden state Temporal feature extraction is performed; after processing by a bidirectional gated loop unit, the last time step is extracted. The output is used as a high-dimensional temporal feature vector. The calculation formula is: At this time, the feature vector The dimension is This feature vector incorporates the contextual temporal dependency information of the microseismic signal and serves as a common input for subsequent steps.

[0039] Time series feature vectors The input to the domain adaptive discriminant network first passes through a gradient reversal layer (GRL). During the forward propagation of training, the GRL keeps the input unchanged. However, during the backpropagation, the GRL multiplies the gradient by a negative coefficient. This information is then passed to the feature extractor, which learns common features that are insensitive to the domain. The features after GRL processing... The input is fed into a domain discrimination network, which consists of: a gradient reversal layer, a fully connected layer, a Dropout layer, a fully connected layer, a LeakyReLU activation function, and finally a domain classification fully connected layer.

[0040] During model training, each training batch consists of equal numbers of source domain samples and target domain samples. The domain label for the source domain samples is denoted as 0, and the domain label for the target domain samples is denoted as 1. The negative coefficient α of the gradient reversal layer is not a fixed value, but is progressively adjusted as training progresses. Where γ is the scheduling rate coefficient, taken as γ=10. p is the training progress, with a value in the range [0,1], and its value is the ratio of the current iteration round to the total number of iteration rounds. In the early stage of training, α is close to 0 to ensure that the classification task is learned first. As training progresses, α gradually increases and tends to 1, and the domain alignment strength gradually increases.

[0041] During adversarial training, the parameters of the domain adaptive discriminant network are updated in the direction that reduces the domain discrimination error, while the parameters of the feature extractor are updated in the direction that increases the domain discrimination error under the action of the gradient reversal layer. The two evolve together in the game, ultimately causing the feature extractor to converge to domain-invariant features that are insensitive to domain information and only retain information related to class discrimination, denoted as . .

[0042] Domain-invariant eigenvectors The signal is input into a microseismic signal classification network to identify the specific category of the microseismic signal and outputs the predicted category probability distribution. eigenvectors The first fully connected layer maps the data to a 64-dimensional feature space, followed by batch normalization, ReLU activation, and Dropout for regularization. The features then pass through a second fully connected layer to reduce the dimension to 32, and are again processed by ReLU activation. Finally, a third fully connected layer maps the data to the class dimension. Output the predicted class probability distribution This distribution is used for subsequent classification loss calculation.

[0043] S103, after amplifying the training samples, the microseismic signal classification model is trained. During training, the model parameters are adjusted and updated to make the model converge and obtain the optimal model weight parameters.

[0044] The selected microseismic event signals and blasting event signals are input into a diffusion enhancement network for model training and sample generation. The constructed class-balanced microseismic signal dataset is divided into training, validation, and test sets in an 8:1:1 ratio. The training set is input into the constructed microseismic signal classification model for training; the validation dataset is used to monitor the training process and adjust hyperparameters; during training, the model parameters are updated using the backpropagation algorithm to achieve model convergence, and the optimal model weight parameters performed on the target domain validation dataset are saved.

[0045] To facilitate model training, this embodiment constructs a composite loss function for optimizing and updating model parameters. The classification loss employs a multi-class cross-entropy loss function to optimize classifier performance, ensuring joint optimization of the classification and domain discrimination tasks. The cross-entropy function expression is as follows: in: For batch size, For the number of categories, For the first The sample at the th The true label of the class, The model predicts the first The sample belongs to the first The probability of a class.

[0046] The domain discriminator uses a binary cross-entropy loss function to optimize the domain adaptation loss. The expression for the binary cross-entropy loss function is as follows: in: For real tags (source domain or target domain). This represents the domain probability predicted by the model.

[0047] To optimize the denoising performance of the diffusion model, the mean squared error (MSE) is used to calculate the noise prediction error of the diffusion model. in: The noise predicted by the diffusion model, This is real noise.

[0048] The aforementioned loss is optimized separately in two training phases: during the training phase of the diffusion enhancement network, the loss is predicted solely using noise. Optimize the parameters of the denoising network; during the training phase of the classification model, use the total loss function... The feature extractor, classifier, and domain adaptive discriminant network are jointly optimized. The classification loss is calculated simultaneously on both source and target domain samples, and the class label of the target domain sample is inherited from its corresponding source domain signal. The weighted hyperparameters control the domain alignment strength.

[0049] In this embodiment, the AdamW optimization algorithm is used to optimize model parameter learning. A Cosine Annealing learning rate scheduler is used to gradually reduce the learning rate during training. The batch size is set to 32, the initial learning rate to 0.01, and the weight decay to 0.001, with 100 iterations. After iterative training to the preset number of training rounds, the optimal model weights are saved. The criterion for determining the optimal model weights is the model's classification performance on the target domain validation dataset (i.e., validation data after injecting noise), ensuring that the saved model matches the actual noisy monitoring conditions.

[0050] Using accuracy and The score is used as the main evaluation metric to objectively evaluate the classification performance of the model. The formula for calculating the accuracy is: in, This represents the number of samples that were actually positive and were predicted as positive by the model. This represents the number of samples that were actually negative but were incorrectly predicted as positive by the model. The formula for calculating fractions is: ,in This represents the number of samples that were actually positive but were incorrectly predicted as negative by the model. TP, FP, and FN are calculated separately for each category, and the precision and F1 score are taken as a macro average for each category.

[0051] S104, Microseismic signal classification is performed using the microseismic signal classification model updated with the optimal model weight parameters.

[0052] The target domain test dataset is input into the microseismic signal classification model loaded with the optimal model weights for testing. During the testing phase, the domain adaptive discriminant module (including the gradient inversion layer and the domain adaptive discriminant network) is completely discarded, and only the bidirectional gated recurrent unit feature extractor and the multi-classifier are retained to perform forward propagation. Noisy test data are constructed according to preset signal-to-noise ratio levels to evaluate the model's classification performance under different noise intensities.

[0053] To evaluate the effectiveness and robustness of the proposed method in real-world industrial scenarios, comprehensive experiments were conducted on the self-built WYMKCT microseismic signal classification dataset. The WYMKCT dataset, derived from real data collected in a coal mine in Shanxi Province, contains 3000 microseismic monitoring signal samples, covering three typical signal categories commonly encountered in deep working conditions: microseismic events, blasting events, and environmental noise. To ensure the standardization of model input, all signals underwent standardized preprocessing to guarantee the same sampling frequency and duration.

[0054] From a set of 3000 microseismic signal samples, microseismic event signals and blasting event signals were selected as training data and input into the diffusion enhancement network for training. The trained diffusion enhancement network was then used to generate additional samples to expand the sample size for specific categories, thereby constructing an enhanced, class-balanced microseismic signal dataset. The enhanced microseismic signal dataset contains a total of 6000 samples, specifically: 2000 microseismic signals, 2000 blasting signals, and 2000 noise signals. The dataset was randomly divided according to a preset ratio of 8:1:1, resulting in a training set of 4800 samples for model training; a validation set of 600 samples for hyperparameter tuning and model selection; and a test set of 600 samples for final performance evaluation and testing. Precision and... The score serves as the primary evaluation metric, while precision is used to measure the accuracy of the model's predictions. The scores are used to comprehensively evaluate the accuracy and recall of the classification model. Table 1 shows the classification performance evaluation results of the method of this invention on the WYMKCT dataset.

[0055] Table 1 Comparison of Classification Performance Evaluation Results As can be seen from Table 1, compared with the existing mainstream time series classification methods, the data augmentation and domain adaptation mechanism of the fusion diffusion model proposed in this application shows significant advantages.

[0056] As can be seen from the above embodiments, this embodiment integrates diffusion enhancement and domain adaptation mechanisms into the microseismic signal classification model. By amplifying the training samples of coal mine microseismic events and blasting events, the number of samples is increased and the sample features are enriched, alleviating the limitation of model training caused by sample scarcity and reducing the risk of overfitting under small sample conditions. Through the domain adaptation mechanism, the influence of signal distribution differences caused by changes in different working conditions and monitoring equipment status on the classification results is reduced, enhancing the model's adaptability to different monitoring conditions. By adjusting and updating the model parameters during training, the model's ability to distinguish between microseismic events and blasting events and its generalization performance are improved, thereby enhancing the accuracy and stability of microseismic signal classification under small sample conditions, and providing signal classification basis for mine seismic source location, energy calculation, and disaster risk assessment.

[0057] Example 2 Corresponding to the above embodiments, this application also provides a microseismic signal classification system with fusion diffusion enhancement and domain adaptation for small samples.

[0058] See Figure 4 The microseismic signal classification system 20 with fusion diffusion enhancement and domain adaptation for small samples provided in this embodiment includes: The data processing module 201 is used to preprocess the acquired raw microseismic monitoring data of the coal mine and extract the microseismic event signals and blasting event signals as training samples.

[0059] Model building module 202 is used to build a microseismic signal classification model that integrates diffusion enhancement and domain adaptation mechanisms; The weight parameter determination module 203 is used to amplify the training samples and then train the microseismic signal classification model. During training, the model parameters are adjusted and updated to make the model converge and obtain the optimal model weight parameters.

[0060] Microseismic signal classification network 204 is used to classify microseismic signals using the updated microseismic signal classification model with optimal model weight parameters.

[0061] It should be noted that the specific content involved in the embodiments of this application can be found in the description of the above method embodiments, and will not be repeated here for the sake of brevity.

[0062] Example 3 Corresponding to the above embodiments, this application also provides an electronic device.

[0063] See Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5As shown, the electronic device 30 may include a processor 301, a memory 302, and a communication unit 303. These components communicate via one or more buses. Those skilled in the art will understand that the electronic device structure shown in the figures does not constitute a limitation on the embodiments of this application. It may be a bus topology or a star topology, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0064] The communication unit 303 is used to establish a communication channel, thereby enabling the electronic device to communicate with other devices.

[0065] The processor 301 serves as the control center of the electronic device, connecting various parts of the device via various interfaces and lines. It executes software programs and / or modules stored in the memory 302, and calls data stored in the memory to perform various functions and / or process data. The processor can be composed of integrated circuits (ICs), such as a single packaged IC or multiple packaged ICs with the same or different functions connected together. For example, the processor 301 may consist only of a central processing unit (CPU). In this embodiment, the CPU may have a single processing core or include multiple processing cores.

[0066] Memory 302 is used to store the execution instructions of processor 301. Memory 302 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0067] When the execution instructions in memory 302 are executed by processor 301, the electronic device 30 is able to perform some or all of the steps in the above method embodiments.

[0068] Example 4 Corresponding to the above embodiments, this application also provides a computer-readable storage medium, wherein the computer-readable storage medium may store a program, wherein when the program runs, it can control the device where the computer-readable storage medium is located to execute some or all of the steps in the above method embodiments. In specific implementation, the computer-readable storage medium may be a magnetic disk, an optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0069] Example 5 Corresponding to the above embodiments, this application also provides a computer program product containing executable instructions that, when executed on a computer, cause the computer to perform some or all of the steps in the above method embodiments.

[0070] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0071] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0072] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0073] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A microseismic signal classification method based on fusion diffusion enhancement and domain adaptation for small samples, characterized in that, include: The acquired raw microseismic monitoring data of the coal mine was preprocessed, and the microseismic event signals and blasting event signals were extracted as training samples. Construct a microseismic signal classification model that integrates diffusion enhancement and domain adaptation mechanisms; After amplifying the training samples, the microseismic signal classification model is trained. During training, the model parameters are adjusted and updated to make the model converge and obtain the optimal model weight parameters. Microseismic signals are classified using the microseismic signal classification model updated with the optimal model weight parameters.

2. The microseismic signal classification method based on fusion diffusion enhancement and domain adaptation for small samples as described in claim 1, characterized in that, The process of preprocessing the acquired raw microseismic monitoring data of the coal mine to extract microseismic event signals and blasting event signals as training samples includes: The raw microseismic monitoring data of the target monitoring area is acquired and preprocessed, including denoising, normalization and signal extraction, to construct a raw sample set containing microseismic signals and their corresponding category labels. Microseismic event signals and blasting event signals were extracted from the original sample set as training samples.

3. The microseismic signal classification method based on fusion diffusion enhancement and domain adaptation for small samples as described in claim 1, characterized in that, The microseismic signal classification model includes: a diffusion enhancement network, a temporal feature extraction network, a domain adaptive discriminant network, and a microseismic signal classification network. The construction of the microseismic signal classification model integrating diffusion enhancement and domain adaptive mechanisms includes: A forward diffusion process is established to gradually inject Gaussian noise into the original signal, and a denoising network conditioned on category labels and time steps is constructed. The forward diffusion process is then combined to form a diffusion enhancement network. Source domain data is constructed based on the signal samples generated by the diffusion enhancement network and the original signal samples, and target domain data is obtained by adding Gaussian white noise to the source domain data. A temporal feature extraction network is constructed using multi-layer stacked bidirectional gated recurrent units. The source domain data and target domain data are connected to the same temporal feature extraction network, and the concatenation result of the forward and backward hidden states is used as the feature output. Construct a domain adaptive discriminant network that includes a gradient reversal layer and a domain discriminant network, and connect it to the output of the temporal feature extraction network to establish a domain adversarial training branch; A multi-layer fully connected network is used to construct a microseismic signal classification network, which is then connected to the output of a temporal feature extraction network. The output dimension of this network is set to the number of signal categories.

4. The microseismic signal classification method based on fusion diffusion enhancement and domain adaptation for small samples as described in claim 3, characterized in that, The process of constructing source domain data based on the signal samples generated by the diffusion enhancement network and the original signal samples, and obtaining target domain data by adding Gaussian white noise to the source domain data, includes: Conditional sampling is performed using a trained denoising network to sample initial noise from a standard Gaussian distribution. By specifying target category conditions and performing iterative denoising using reverse diffusion, realistic synthetic microseismic and blast signals are generated. in: This is the signal after denoising in the previous step. The current noise signal, For time steps, For target category conditions, The cumulative coefficient is... To predict noise components, The noise standard deviation of the reverse process. Standard Gaussian noise; The generated samples are injected into the original dataset to construct a class-balanced augmented dataset as the source domain data; Additive white Gaussian noise is injected into the source domain data to form the target domain data, which is used to simulate noisy working conditions in actual monitoring.

5. The microseismic signal classification method for small sample fusion diffusion enhancement and domain adaptation according to claim 4, characterized in that, The temporal feature extraction network is constructed using multi-layered stacked bidirectional gated recurrent units. The source domain data and target domain data are connected to the same temporal feature extraction network, and the concatenation result of the forward and backward hidden states is used as the feature output, including: A multi-layer bidirectional gated loop unit is constructed for timing modeling, and source domain data and target domain data are input into the multi-layer bidirectional gated loop unit; The bidirectional gated loop unit is configured as a multi-layer stacked structure, at each time step The bidirectional gated loop unit calculates the forward hidden state respectively. and backward hidden state ; After processing by the bidirectional gated recurrent unit, the output of the last time step is extracted as a high-dimensional temporal feature vector. The calculation formula is: .

6. The microseismic signal classification method based on fusion diffusion enhancement and domain adaptation for small samples as described in claim 5, characterized in that, The construction of a domain adaptive discriminant network, comprising a gradient reversal layer and a domain discriminant network, and its connection to the output of a temporal feature extraction network, establishes a domain adversarial training branch, including: The First, the gradient inversion layer (GRL) processes the feature extractor, enabling it to learn common features that are insensitive to the domain. The features processed by GRL are input into a domain adaptive discriminant network, which consists of: a fully connected layer, a LeakyReLU activation function, a Dropout layer, a fully connected layer, a LeakyReLU activation function, and a final domain classification fully connected layer. During adversarial training, the parameters of the domain adaptive discriminant network are updated in the direction that reduces the domain discrimination error, while the parameters of the feature extractor are updated in the direction that increases the domain discrimination error under the action of the gradient reversal layer. Ultimately, the feature extractor converges to domain-invariant features that are insensitive to domain information and only retain information related to class discrimination. .

7. The microseismic signal classification method for small sample fusion diffusion enhancement and domain adaptation according to claim 3, characterized in that, After amplifying the training samples, the microseismic signal classification model is trained. During training, the model parameters are adjusted and updated to bring the model to converge and obtain the optimal model weight parameters, including: The training samples were amplified and then divided into a training set, a validation set, and a test set. The microseismic signal classification model is trained using the training set. The model parameters are adjusted using the validation set. The model parameters are updated using the backpropagation algorithm to bring the model to converge and obtain the optimal model weight parameters.

8. The microseismic signal classification method for small sample fusion diffusion enhancement and domain adaptation according to claim 7, characterized in that, Also includes: Construct a composite loss function for optimizing and updating model parameters, with the classification loss employing the multi-class cross-entropy loss function. The domain discriminator uses a binary cross-entropy loss function to optimize the domain adaptation loss. The noise prediction loss of the diffusion model is calculated using the mean square error loss. ; During the training phase of the diffusion-enhanced network, with Optimize the parameters of the denoising network; during the training phase of the classification model, jointly optimize the feature extractor, classifier, and domain adaptive discriminant network using the total loss function; The model training parameters are set, and the AdamW optimization algorithm is used to optimize the model parameter learning. During the training process, the learning rate is gradually reduced, and after iterative training to the preset number of training rounds, the optimal model weights are saved.

9. The microseismic signal classification method for small sample fusion diffusion enhancement and domain adaptation according to claim 8, characterized in that, The expression for the total loss function is: Multi-class cross-entropy loss function for: The binary cross-entropy loss function is used to optimize the domain adaptation loss. for: Noise prediction loss for: in: The weighted hyperparameters for controlling the alignment strength of the control domain, For batch size, For the number of categories, For the first The sample at the th The true label of the class, The model predicts the first The sample belongs to the first The probability of a class; For real labels, The domain probability predicted by the model; The noise predicted by the diffusion model, This is real noise.

10. The microseismic signal classification method for small sample fusion diffusion enhancement and domain adaptation according to claim 7, characterized in that, The microseismic signal classification using the microseismic signal classification model updated with the optimal model weight parameters includes: The test set is input into a microseismic signal classification model loaded with the optimal model weights for testing; During the testing phase, the domain adaptive discrimination module was completely discarded, and only the bidirectional gated recurrent unit feature extractor and multi-classifier were retained to perform forward propagation. Noise-added test data were constructed according to the preset signal-to-noise ratio levels to evaluate the classification performance of the model under different noise intensities.

Images