Micro-seismic data vacancy filling method, system and equipment
By using the GMM-Transformer model to train and complete the microseismic monitoring data in microseismic monitoring technology, the problems of missing data and insufficient completion accuracy in complex engineering environments are solved, efficient and accurate microseismic data completion are achieved, and engineering safety is enhanced.
Patent Information
- Application Number
- CN202510207908.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-25
- Publication Date
- 2025-06-10
AI Technical Summary
The existing microseismic monitoring technology has problems such as missing data, insufficient completion accuracy and high algorithm complexity in complex engineering environments, which affect the accuracy and reliability of disaster warnings.
The GMM-Transformer model is used to train and complete the microseismic monitoring data. The GMM model is used to learn spatial feature distribution laws and the Transformer model is used to learn timing feature evolution laws, and the data is completed by combining the output of the dual model.
It improves the accuracy and reliability of microseismic data completion, can effectively solve the problems of data loss and insufficient completion accuracy, reduces the computational complexity of the algorithm, and enhances the safety of deep buried projects.
Smart Images
Figure CN120123649A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of geomechanics, and particularly to a method, system, and device for filling in missing microseismic data. Background Art
[0002] Currently, in deep-buried engineering, microseismic monitoring technology has become a key means of monitoring the stability of rock masses and warning of potential disasters. However, despite the significant progress made in improving safety through microseismic monitoring technology, its application still faces some technical challenges and deficiencies.
[0003] Firstly, the coverage of microseismic monitoring equipment is usually limited, and it has extremely high requirements for accuracy and sensitivity. This requires the equipment to be adjusted according to the specific monitoring area to ensure the reliability and integrity of the data. The monitoring equipment often needs to manually adjust the sensor array, and this operation is prone to causing the loss of some data, affecting the accuracy of disaster warning. In addition, the sensors and interpreters of microseismic monitoring equipment are usually connected by physical lines, and are easily affected by human interference, equipment failures, or external environmental changes during the construction process, resulting in the interruption or loss of monitoring data. These missing data often lead to the inability to capture potential disaster risks in a timely manner, and may even cause delays or errors in disaster warning, posing a major hidden danger to construction safety.
[0004] Regarding the problem of missing data, traditional data completion methods, such as those based on statistical models and interpolation algorithms (such as KNN interpolation method and Gaussian mixture model GMM), although can speculate missing data to a certain extent, still have great limitations. When dealing with multi-parameter (such as time, energy, etc.) microseismic data, these methods are difficult to consider the correlation between these features, resulting in insufficient accuracy of the completion results. Especially in the processing of multi-dimensional complex data, traditional methods cannot fully demonstrate their advantages.
[0005] With the development of artificial intelligence technology, deep learning methods have been gradually introduced into the field of microseismic data completion. Methods such as long short-term memory network (LSTM), support vector machine (SVM), and random forest (RF) have made certain progress, and these methods perform relatively well in the prediction and learning of high-dimensional data. However, they still have certain deficiencies. For example, LSTM is prone to problems of gradient disappearance or explosion when dealing with long time series data, SVM has a high computational complexity and consumes a large amount of computing resources when dealing with large-scale data, and RF performs poorly when dealing with long-term dependent data. Summary of the Invention
[0006] To solve the above problems, the present application provides a microseismic data filling method, system and device to improve the accuracy and reliability of microseismic data filling, aiming to solve the problems of data loss, insufficient filling accuracy and high algorithm complexity in the existing microseismic monitoring technology. Its application in complex engineering environments will provide strong guarantee for the safety of various deep-buried projects and underground projects. The technical solutions are as follows:
[0007] The present application provides a microseismic data filling method, including the following steps: S1 Data sample division: Divide the microseismic monitoring data into a training sample set and a filling sample set according to the blasting nodes; S2 Train the training sample set through the GMM-Transformer model and extract the rules; S3 Import the rules extracted in S2 into the filling sample set, and predict and fill the missing data in the filling samples through the GMM-Transformer model; S4 Output the data filling result and generate a complete microseismic monitoring data sequence.
[0008] For example, in the microseismic data filling method provided in an embodiment, in S1, the microseismic monitoring data is sampled based on the blasting nodes, and the samples with the maximum time difference between two adjacent microseismic events less than 3 hours are defined as training samples, and the samples with the time difference greater than 3 hours are defined as filling samples.
[0009] For example, in the microseismic data filling method provided in an embodiment, in S2, the training sample set is trained through the GMM-Transformer model, where: the GMM model is used to learn and extract the distribution rules of the spatial features of microseismic events; the Transformer model is used to learn and extract the temporal evolution rules of rupture time, number of microseismic events, seismic moment, static stress drop and microseismic release energy.
[0010] For example, in the microseismic data filling method provided in an embodiment, the Transformer model is used to capture the dynamic features of microseismic events, specifically including the following steps: encode the time series data, where the time series data includes: seismic moment, microseismic release energy and static stress drop; capture the long-term dependencies in the time series through the self-attention mechanism; predict the dynamic features of the missing events based on the encoding results.
[0011] For example, in the microseismic data filling method provided in an embodiment, it also includes defining a training window: splitting the microseismic events based on the time nodes of blasting activities and establishing corresponding time windows to learn the evolution rules of each feature of the microseismic data.
[0012] For example, in the microseismic data filling method provided in an embodiment, in step S3, the method for constructing the GMM-Transformer data filling model includes the following steps: fusing the outputs of the GMM model and the Transformer model; further processing the fused features through a fully connected layer; and outputting the completed microseismic data.
[0013] For example, in the microseismic data filling method provided in an embodiment, the method is applicable to microseismic monitoring in deep-buried projects, including tunnel projects, mine projects, and underground projects.
[0014] The second aspect of this application provides a microseismic data filling system based on a GMM-Transformer model, including the following modules: a blasting node parsing module: used to extract blasting time nodes from construction logs and divide the training sample set and the filling sample set; a dual-model collaborative training module: for the training sample set, the GMM and Transformer models are respectively run to output spatial feature laws and dynamic feature laws; a law migration engine: injecting the training laws into the filling sample set to drive the generation of the GMM-Transformer prediction model.
[0015] The third aspect of this application provides a computer device, including a memory, a processor, and a computer program stored on the memory. When the processor executes the program, the steps of the above method are implemented.
[0016] The beneficial effects brought by a microseismic data filling method, system, and device provided in some embodiments of this application are as follows: By combining the advantages of the GMM and Transformer models, this application realizes high-precision filling of missing microseismic data, can effectively solve the problems of data loss and insufficient filling accuracy in existing microseismic monitoring technologies, can effectively supplement the missing microseismic data during the monitoring process, has high practical value, and provides strong technical support for the microseismic monitoring and early warning system in complex engineering projects such as tunnel projects. Description of the Drawings
[0017] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.
[0018] Figure 1 It is a flowchart of the microseismic data filling method of this application. Detailed Embodiments
[0019] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.
[0020] Unless otherwise defined, the technical terms or scientific terms used in this disclosure shall have the ordinary meaning understood by those of ordinary skill in the art to which this disclosure pertains. The "first", "second" and similar terms used in this disclosure do not denote any order, quantity or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or objects appearing before this term cover the elements or objects listed after this term and their equivalents, without excluding other elements or objects. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left", "right" are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.
[0021] When dealing with microseismic data in a complex rock mass environment, there are still multiple problems in existing microseismic monitoring technologies, such as missing monitoring data, insufficient filling accuracy, high algorithm complexity, etc. Traditional methods based on statistical models and interpolation algorithms (such as KNN interpolation method and Gaussian mixture model GMM) are effective in dealing with spatial distribution problems, but their accuracy and adaptability are limited when facing multi-dimensional and complexly correlated microseismic data. Existing deep learning methods, such as long short-term memory network (LSTM) and support vector machine (SVM), although have certain advantages in dealing with time series data, are prone to the problem of gradient disappearance when dealing with long time series, and have a high computational complexity. Random forest (RF) performs excellently in dealing with local features, but its effect in capturing long-term dependence relationships is poor.
[0022] In view of this, based on the GMM-Transformer model, the present application proposes a microseismic data filling method to solve the problems that the current microseismic monitoring technology cannot ensure the integrity and accuracy of monitoring data in a complex engineering environment, affecting the timeliness and reliability of disaster warning; the current traditional data filling methods cannot effectively capture the complex correlations between various features when dealing with multi-parameter microseismic data, resulting in unsatisfactory filling effects; and the current deep learning methods have problems such as high computational complexity and gradient disappearance when dealing with large-scale data and long time series data, as Figure 1 shown, including the following steps:
[0023] S1 Data sample division: Divide the microseismic monitoring data into a training sample set and a supplementary sample set according to the blasting nodes;
[0024] Specifically, based on the blasting nodes, divide the microseismic monitoring data into samples. Define the samples with a maximum time difference less than 3 hours between two adjacent microseismic events as training samples, that is, Figure 1 in the training set, and define the samples with a time difference greater than 3 hours as supplementary samples, that is, Figure 1 in the prediction set.
[0025] Microseismic missing data refers to the events missing when the time difference between two consecutive events monitored by the monitoring device exceeds a certain value. Different engineering environments and equipment failure problems lead to different degrees of monitoring data loss. For example, the events required for moving sensors in drill-and-blast tunnels are generally 6 - 8 hours, while for TBM tunnels, it is 3 - 5 hours. The time required to repair the line damage caused by on-site construction needs to be determined according to the location of the line damage, generally 1 - 3 hours. If this time interval is too small, it will increase the training cost of the model and lead to overfitting of the data. If it is too large, the model will have insufficient learning of the evolution law of each feature of the microseismic data within the time window, thus affecting the prediction effect of the missing values. Considering the above effects comprehensively, this application defines the samples with a time difference exceeding 3 hours between two adjacent microseismic events as microseismic missing data samples.
[0026] S2 Train the training sample set through the GMM-Transformer model and extract the rules;
[0027] This step involves the selection of microseismic data characteristic parameters: The rock mass rupture information monitored by the microseismic monitoring device through sensors mainly includes the time when the rupture event occurs, three-dimensional spatial coordinates, seismic moment, microseismic release energy, static stress drop, dynamic stress drop, and P-wave and S-wave radiation energy, etc. In the selection of training characteristic parameters, if all microseismic characteristics are considered for training, learning, and supplementation, it will undoubtedly increase the training cost, which is not only reflected in higher computing resource consumption and longer training time, but may also lead to overfitting of the model and reduce its generalization ability on unseen data. In addition, including too many redundant or low-correlation features will increase the complexity of data processing, reduce the training efficiency of the model, and may introduce noise, affecting the accuracy of the supplemented data. Existing research shows that the number of microseismic events, microseismic release energy, and apparent volume of microseismic events in microseismic monitoring data are three basic, independent variables that can fully characterize the characteristics of rupture events. Since the apparent volume of microseismic events can be calculated from the seismic moment and static stress drop, in the data supplementation work, the number of microseismic events, seismic moment, static stress drop, and microseismic release energy can be used as input characteristic parameters for training and supplementation.
[0028] Among them, spatial position information is the core feature of rock rupture events, which can help describe the distribution characteristics of microseismic events in underground space. Since the spatial distribution involves more complex multidimensional data, the use of Gaussian mixture model (GMM) to predict spatial position information can better capture the potential pattern of its spatial distribution; seismic moment, microseismic release energy and static stress drop are dynamic characteristics of rock rupture, which involve the analysis of time series data. In order to more accurately capture the changing laws of these parameters, the Transformer model is used for prediction.
[0029] Transformer can effectively handle long dependencies in time series through the self-attention mechanism. It is suitable for capturing complex time series characteristics in microseismic data, especially in long-span data analysis. It can fully explore the correlation between different time nodes and improve prediction accuracy.
[0030] S3 imports the rules extracted by S2 into the missing sample set, and predicts and completes the missing data in the missing samples through the GMM-Transformer model;
[0031] Specifically, the method for constructing the GMM-Transformer data filling model includes the following steps:
[0032] Fusion of the outputs of the GMM model and the Transformer model;
[0033] The fused features are further processed through the fully connected layer;
[0034] Output the completed microseismic data.
[0035] S4 outputs the data filling results to generate a complete microseismic monitoring data sequence.
[0036] In order to enable the GMM-Transformer model to better learn the evolution laws of various features of microseismic data, microseismic events can be split based on the time nodes of blasting activities and corresponding time windows can be established to train and fill in the gaps of various features of microseismic data.
[0037] Compared with the prior art, the microseismic data filling method based on the GMM-Transformer model in this application has significant advantages. First of all, by combining the advantages of GMM and Transformer, this application overcomes the deficiencies of traditional methods in processing multi-parameter microseismic data. GMM is used to effectively predict the spatial location information of microseismic events and can capture the complex spatial distribution characteristics of the data. While Transformer can handle the long-term dependencies in multi-dimensional data through the self-attention mechanism, solving the problems of gradient disappearance and computational complexity of LSTM and SVM in processing large-scale data. Therefore, the GMM-Transformer model can effectively improve the accuracy and reliability of microseismic data filling while ensuring a relatively low computational cost.
[0038] In addition, this application shows high practicality in actual applications. The model can accurately recover the missing data, improve the accuracy of rock mass stability assessment, provide more comprehensive and accurate information for disaster warning, and thus effectively prevent engineering disasters such as rock bursts and collapses. Compared with traditional methods, this application provides a more stable and accurate data filling solution, improves the practicality of the microseismic monitoring system in complex engineering environments, and greatly enhances the safety of deep-buried projects such as tunnels.
[0039] The microseismic data filling method of this application is applicable to microseismic monitoring in deep-buried projects, including tunnel projects, mine projects, and underground projects.
[0040] The second aspect of this application provides a microseismic data filling system based on the GMM-Transformer model, including the following modules:
[0041] Blasting node analysis module: used to extract the blasting time nodes from the construction log and divide the training sample set and the filling sample set;
[0042] Dual-model collaborative training module: For the training sample set, run the GMM and Transformer models respectively, and output the spatial feature law and the dynamic feature law;
[0043] Law migration engine: Inject the training law into the filling sample set to drive the generation of the GMM-Transformer prediction model.
[0044] The third aspect of this application provides a computer device, including a memory, a processor, and a computer program stored on the memory. When the processor executes the program, it implements the steps of the above method.
[0045] Although the embodiments of the present application have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be fully applied to various fields suitable for the present application. For those skilled in the art, additional modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present application is not limited to the specific details and the examples shown and described herein.
Claims
1. A method for filling gaps in microseismic data, characterized in that: The following steps are involved: S1 Data sample division: The microseismic monitoring data is divided into training sample sets and gap-filling sample sets according to blasting nodes; S2 trains the training sample set through a GMM-Transformer model and extracts rules; S3 imports the rules extracted by S2 into the missing sample set, and predicts and completes the missing data in the missing samples through the GMM-Transformer model; S4 outputs the data filling results to generate a complete microseismic monitoring data sequence.
2. The microseismic data gap filling method according to claim 1, characterized in that: In S1, the microseismic monitoring data is divided into samples based on the blasting nodes, and the samples with a maximum time difference of less than 3 hours between two adjacent microseismic events are defined as training samples, and the samples with a time difference of more than 3 hours are defined as filling samples.
3. The microseismic data gap filling method according to claim 2, characterized in that: In S2, the training sample set is trained by a GMM-Transformer model, wherein: The GMM model is used to learn and extract the distribution law of the spatial characteristics of microseismic events; The Transformer model is used to learn and extract the temporal evolution of rupture time, number of microseismic events, seismic moment, static stress drop and microseismic release energy.
4. The microseismic data gap filling method according to claim 3, characterized in that: The Transformer model is used to capture the dynamic characteristics of microseismic events, and specifically includes the following steps: Encode the time series data, including seismic moment, microseismic release energy and static stress drop; Capturing long-term dependencies in time series through self-attention mechanism; Predicting the dynamic characteristics of missing events based on encoding results.
5. The microseismic data gap filling method according to claim 1, characterized in that: It also includes defining training windows: splitting microseismic events based on the time nodes of blasting activities and establishing corresponding time windows to learn the evolution laws of various characteristics of microseismic data.
6. The microseismic data gap filling method according to claim 1, characterized in that: In S3, the method for constructing the GMM-Transformer data filling model includes the following steps: Fusion of the outputs of the GMM model and the Transformer model; The fused features are further processed through the fully connected layer; Output the completed microseismic data.
7. The microseismic data gap filling method according to claim 1, characterized in that: The method is suitable for microseismic monitoring in deep buried engineering, including tunnel engineering, mining engineering and underground engineering.
8. A microseismic data gap filling system based on the GMM-Transformer model, characterized in that: Includes the following modules: Blasting node parsing module: used to extract blasting time nodes from construction logs and divide them into training sample sets and gap-filling sample sets; Dual-model collaborative training module: for the training sample set, run the GMM and Transformer models respectively to output spatial feature laws and dynamic feature laws; Rule transfer engine: Injects training rules into the missing sample set to drive the generation of the GMM-Transformer prediction model.
9. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory, wherein when the processor executes the program, the steps of the method according to any one of claims 1 to 7 are implemented.