An early warning method based on the tendency of microseismic energy accumulation

Abstract

This application proposes an early warning method based on the tendency of microseismic energy accumulation, belonging to the field of seismic data processing. The method includes: acquiring a historical microseismic log set from a seismic monitoring sensor array in a coal mine area; performing phased processing with timestamps to obtain a set of microseismic energy feature subsequences; performing first-order difference operations to construct a set of differential energy feature subsequences; using a spatiotemporal guidance module to perform trend correlation analysis to obtain a set of trend features; constructing an energy accumulation tendency analyzer based on the trend feature set; acquiring real-time microseismic logs, combining the analyzer for anomaly identification, and triggering an early warning command. This application solves the technical problem in existing technologies where the dynamic change trend of microseismic energy cannot be accurately captured, leading to insufficient accuracy in energy accumulation early warning. It achieves the technical effect of improving the accuracy and timeliness of microseismic energy accumulation early warning through differential energy feature analysis and spatiotemporal trend correlation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of seismic data processing, and in particular to a microseismic energy accumulation tendency-based early warning method. BACKGROUND

[0002] During the process of coal mining, microseismic phenomena often occur due to the redistribution of geological stress and the change of rock mass structure. The abnormal accumulation of microseismic energy is often an important precursor of coal mine disasters, so effective monitoring and early warning of the microseismic energy accumulation state is of great significance to ensure the safety of coal mine production.

[0003] Currently, the existing microseismic monitoring and early warning technology mainly analyzes the static characteristics such as energy amplitude and frequency of microseismic events. For example, by setting an energy threshold or event frequency threshold to determine the abnormal state, or using statistical methods to analyze the distribution characteristics of microseismic energy. These methods can identify microseismic anomalies to some extent, but have obvious technical defects. Specifically, the existing technology mainly focuses on the absolute numerical characteristics of microseismic energy, while ignoring the dynamic process of energy change. The accumulation of microseismic energy is a gradual dynamic process, and its change trend and rate often reflect the proximity of disasters better than the absolute numerical value. However, the existing methods lack the ability to accurately capture the dynamic change trend of microseismic energy, and cannot effectively identify the tendency characteristics of energy accumulation, resulting in insufficient accuracy of energy accumulation early warning, and prone to false positives or false negatives, which is difficult to meet the actual needs of coal mine safety production. SUMMARY

[0004] The present application provides a microseismic energy accumulation tendency-based early warning method to solve the technical problem of the existing technology that cannot accurately capture the dynamic change trend of microseismic energy, resulting in insufficient accuracy of energy accumulation early warning.

[0005] The technical solution of the present application to solve the above technical problems is as follows:

[0006] In a first aspect, the present application provides a microseismic energy accumulation tendency-based early warning method, comprising: obtaining a set of historical microseismic logs collected by a seismic monitoring sensor array arranged in a target coal mine region within a historical time window, wherein each historical microseismic log comprises a log timestamp; performing phased processing on the set of historical microseismic logs in combination with the log timestamps to obtain a set of historical microseismic energy feature subsequences; performing first-order difference operation on the set of historical microseismic energy feature subsequences respectively to construct a set of historical differential energy feature subsequences; performing trend correlation analysis on the set of historical microseismic energy feature subsequences and the set of historical differential energy feature subsequences respectively by using a space-time oriented module to obtain a set of historical microseismic energy trend features and a set of historical differential energy trend features; constructing an energy accumulation tendency analyzer based on the set of historical microseismic energy trend features and the set of historical differential energy trend features; obtaining a set of real-time microseismic log sequences obtained by real-time monitoring of the seismic monitoring sensor array, and performing abnormality identification in combination with the energy accumulation tendency analyzer to trigger a microseismic energy accumulation early warning instruction.

[0007] Optionally, the phased processing on the set of historical microseismic logs in combination with the log timestamps to obtain a set of historical microseismic energy feature subsequences comprises: performing energy feature extraction on the set of historical microseismic logs by using a pre-constructed feature extractor to obtain a set of historical microseismic energy features; performing sequence processing on the set of historical microseismic energy features according to the log timestamps to obtain a historical microseismic energy feature sequence; and slicing the historical microseismic energy feature sequence according to a preset sliding window to determine the set of historical microseismic energy feature subsequences.

[0008] Optionally, the slicing of the historical microseismic energy feature sequence according to a preset sliding window to determine the set of historical microseismic energy feature subsequences comprises: slicing the historical microseismic energy feature sequence according to a preset sliding window to obtain an initial set of historical microseismic energy feature subsequences; and performing adjacent fusion analysis on the initial set of historical microseismic energy feature subsequences to obtain the set of historical microseismic energy feature subsequences.

[0009] Optionally, the trend correlation analysis on the set of historical microseismic energy feature subsequences and the set of historical differential energy feature subsequences respectively by using a space-time oriented module to obtain a set of historical microseismic energy trend features and a set of historical differential energy trend features comprises: performing supervised training on a framework constructed based on a feedforward neural network based on a preset time scale and a sample time-oriented training set to obtain a time-oriented submodule; performing supervised training on a framework constructed based on a feedforward neural network based on a preset space scale and a sample space-oriented training set to obtain a space-oriented submodule; and connecting the time-oriented submodule and the space-oriented submodule in parallel to obtain the space-time oriented module.

[0010] Optionally, the method further comprises: performing time-space feature extraction on the historical microseismic energy feature sequence and the historical differential energy feature sequence respectively by using the time-oriented sub-module and the space-oriented sub-module to obtain historical microseismic energy time trend features, historical microseismic energy space trend features, historical differential energy time trend features, and historical differential energy space trend features; performing trend correlation analysis on the historical microseismic energy time trend features and the historical microseismic energy space trend features to obtain historical microseismic energy trend features; and performing trend correlation analysis on the historical differential energy time trend features and the historical differential energy space trend features to obtain historical differential energy trend features.

[0011] Optionally, the trend correlation analysis on the historical microseismic energy time trend features and the historical microseismic energy space trend features to obtain historical microseismic energy trend features comprises: calculating feature similarity of the historical microseismic energy time trend features and the historical microseismic energy space trend features to obtain a feature similarity set; performing normalization processing on the feature similarity set and matrixing the processing result to obtain a historical microseismic energy time adjacency matrix; and performing convolution on the historical microseismic energy time adjacency matrix and the historical microseismic energy space trend features to obtain the historical microseismic energy trend features.

[0012] Optionally, the construction of the energy accumulation tendency analyzer based on the historical microseismic energy trend feature set and the historical differential energy trend feature set comprises: taking each historical microseismic energy trend feature and historical differential energy trend feature in the historical microseismic energy trend feature set and the historical differential energy trend feature set as a set of positive sample pairs to obtain a positive sample pair set; mapping the historical microseismic energy trend feature set and the historical differential energy trend feature set into historical microseismic energy trend feature probability distribution and historical differential energy trend feature probability distribution respectively; performing symmetric KL divergence calculation on the historical microseismic energy trend feature probability distribution and the historical differential energy trend feature probability distribution to determine a consistency loss function; and performing supervised calculation on a framework constructed based on a convolutional neural network by taking the positive sample pair set as a comparative representation learning input, and constructing the energy accumulation tendency analyzer by minimizing the consistency loss function.

[0013] Optionally, the mapping of the historical microseismic energy trend feature set and the historical differential energy trend feature set into historical microseismic energy trend feature probability distribution and historical differential energy trend feature probability distribution respectively comprises: performing normalization and softmax mapping on the historical microseismic energy trend feature set to obtain the historical microseismic energy trend feature probability distribution; and performing normalization and softmax mapping on the historical differential energy trend feature set to obtain the historical differential energy trend feature probability distribution.

[0014] Optionally, the acquiring the real-time microseismic log sequence set obtained by the seismic monitoring sensor array in real time, in combination with the energy accumulation tendency analyzer, performing anomaly identification, triggering a microseismic energy accumulation early warning instruction, comprises: performing energy feature extraction on the real-time microseismic log sequence set to determine a real-time microseismic energy feature sequence set; performing first-order difference operation on the real-time microseismic energy feature sequence set to determine a real-time difference energy feature sequence set; performing trend correlation analysis on the real-time microseismic energy feature sequence set and the real-time difference energy feature sequence set respectively by using a space-time orientation module to determine a real-time microseismic energy trend feature set and a real-time difference energy trend feature set; performing consistency loss value analysis on the real-time microseismic energy trend feature set and the real-time difference energy trend feature set by using the energy accumulation tendency analyzer to determine a consistency loss value set; and analyzing the consistency loss value set, and if there is an anomaly, triggering a microseismic energy accumulation early warning instruction.

[0015] Optionally, the consistency loss value analysis on the real-time microseismic energy trend feature set and the real-time difference energy trend feature set by using the energy accumulation tendency analyzer to determine a consistency loss value set comprises: acquiring a monitoring coordinate corresponding to a consistency loss value greater than or equal to a preset consistency loss value threshold in the consistency loss value set to obtain an abnormal monitoring coordinate set; performing near-neighbor region division on the abnormal monitoring coordinate set according to a preset monitoring coordinate near-neighbor threshold to obtain K division near-neighbor regions, wherein K is a positive integer; traversing and counting the number of abnormal monitoring coordinates in the K division near-neighbor regions, and obtaining a K division near-neighbor region abnormal coefficient by placing the statistical result in the area of the K division near-neighbor regions; and when any one of the K division near-neighbor region abnormal coefficients is greater than or equal to a preset division near-neighbor region abnormal coefficient threshold, triggering a microseismic energy accumulation early warning instruction.

[0016] The beneficial effects of the present application are:

[0017] The historical microseismic log set collected by the seismic monitoring sensor array arranged in the target coal mine area in a historical time window is acquired, wherein each historical microseismic log includes a log timestamp, thereby establishing a complete historical microseismic data basis to provide data support for subsequent energy feature analysis; the historical microseismic log set is processed in stages in combination with the log timestamp, and a historical microseismic energy feature subsequence set is obtained, thereby segmenting continuous historical data in the time dimension to facilitate capturing energy change rules in different periods; first-order difference operation is respectively performed on the historical microseismic energy feature subsequence set to construct a historical differential energy feature subsequence set, thereby converting absolute energy values into relative change amounts to highlight dynamic features and change rates of energy change; trend correlation analysis is respectively performed on the historical microseismic energy feature subsequence set and the historical differential energy feature subsequence set by using a space-time oriented module to obtain a historical microseismic energy trend feature set and a historical differential energy trend feature set, thereby mining correlation rules of energy change from two dimensions of time and space to identify trend features of energy accumulation; an energy accumulation tendency analyzer is constructed based on the historical microseismic energy trend feature set and the historical differential energy trend feature set to provide analysis decision basis for real-time monitoring; a real-time microseismic log sequence set obtained by real-time monitoring of the seismic monitoring sensor array is acquired, and abnormality is identified in combination with the energy accumulation tendency analyzer to trigger a microseismic energy accumulation early warning instruction to realize real-time analysis and judgment of a current microseismic state, and an early warning is timely issued when an energy accumulation abnormal tendency is detected.

[0018] Through the above technical solution, by capturing dynamic change features of microseismic energy, trend correlation analysis is performed in combination with a space-time oriented module to construct an early warning mechanism based on energy accumulation tendency, thereby effectively solving the technical problem that the prior art cannot accurately capture dynamic change trends of microseismic energy, and achieving the technical effects of improving microseismic energy accumulation early warning accuracy and timeliness. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A flowchart of an early warning method based on microseismic energy accumulation tendency provided by the present application is shown in the figure.

[0020] Figure 2 A flowchart of triggering a microseismic energy accumulation early warning instruction provided by the present application is shown in the figure. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0022] In the description of the present application, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can be explicitly or implicitly included one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly specified.

[0023] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or description". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given in order to enable any person skilled in the art to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that those skilled in the art can recognize that the present application can be implemented without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope consistent with the principles and characteristics disclosed.

[0024] As Figure 1 shown, the embodiment of the present application provides a warning method based on microseismic energy accumulation tendency, comprising:

[0025] S1, obtaining a set of historical microseismic logs collected by a seismic monitoring sensor array arranged in a target coal mine area within a historical time window, wherein each historical microseismic log includes a log timestamp.

[0026] Specifically, the target coal mine area refers to a specific coal mining area that needs to be warned of microseismic energy accumulation, which includes the working face being mined or about to be mined and its surrounding influence range, specifically covering the coal mining face, the tunneling roadway, the goaf and the surrounding rock area that may be affected by mining disturbance. A seismic monitoring sensor array is pre-arranged in the target coal mine area, which is composed of several microseismic monitoring sensors distributedly installed at different positions in the coal mine, such as key areas like main roadways, working face vicinity, goaf edge, etc. Each sensor is equipped with a high-precision vibration detection element and a data processing unit, which can monitor the microseismic signals generated by the coal and rock mass during mining in real time, and convert the detected vibration data into digital signals for recording and transmission.

[0027] The historical time window refers to a specific time range for obtaining historical microseismic logs, and the length of the time window is set according to the geological conditions, mining progress and historical microseismic activity law of the coal mine. For example, for a working face that is being mined, the historical time window can be set to the time period from 3 months before the working face is mined to the current time, to ensure that enough historical microseismic logs are obtained for subsequent analysis. Within this historical time window, the seismic monitoring sensor array continues to work, and when a vibration signal exceeding a preset threshold is detected, the data recording process is automatically triggered. Each historical microseismic log is a complete digital record of a single microseismic event, containing a log timestamp of the microseismic event. The log timestamp is a required field of each historical microseismic log, which accurately records the specific time of the microseismic event, and the format is usually "year-month-day hour: minute: second.millisecond", such as "2024-03-15 14:25:36.123". In addition to the log timestamp, each historical microseismic log also contains information such as magnitude, hypocenter coordinates (x, y, z), energy release, and detection sensor number. These historical microseismic logs are stored in chronological order to form a historical microseismic log set, providing a data basis for subsequent microseismic energy feature extraction and accumulation tendency analysis.

[0028] S2, in combination with the log timestamp, the historical microseismic log set is processed in stages to obtain a historical microseismic energy feature subsequence set.

[0029] Specifically, the obtained historical microseismic log set is sequentially organized and segmented according to the time dimension to extract feature subsequences reflecting the microseismic energy change law. The staged processing refers to arranging the historical microseismic log set in time sequence according to the log timestamp, and then dividing the entire historical time window into multiple continuous time stages according to a preset time period length, each time stage corresponding to a historical microseismic energy feature subsequence.

[0030] First, according to the log timestamp in each historical microseismic log, the historical microseismic log set is sorted in chronological order to form an ordered sequence in ascending order of time. Then, a preset stage time length is set, which can be adjusted according to the periodic characteristics of microseismic activity and warning requirements, and typical values are 1 hour, 6 hours or 24 hours. Then, the entire historical time window is divided into several continuous and non-overlapping time periods with the stage time length as the interval, and all historical microseismic logs contained in each time period form a processing stage.

[0031] For the historical microseismic log in each processing stage, its microseismic energy related parameters are extracted, including magnitude size, energy release amount, focal depth and other information, and statistical characteristics in this stage are calculated, such as average energy, maximum energy, energy fluctuation variance, microseismic event frequency, etc. These statistical characteristics are combined in the order of time stages to form a subsequence reflecting the time sequence change characteristics of historical microseismic energy, that is, a historical microseismic energy feature subsequence. Since the entire historical time window is divided into multiple processing stages, multiple historical microseismic energy feature subsequences are finally obtained, constituting a historical microseismic energy feature subsequence set, which provides structured data for subsequent energy accumulation tendency analysis.

[0032] S3, performing first-order difference operation on the historical microseismic energy feature subsequence set respectively to construct a historical differential energy feature subsequence set.

[0033] Specifically, the change trend and fluctuation law of the historical microseismic energy feature are captured through the first-order difference operation, thereby constructing a historical differential energy feature subsequence set that can reflect the energy accumulation tendency. Wherein, the first-order difference operation refers to the subtraction operation of adjacent two data points in the historical microseismic energy feature subsequence set to obtain the change rate information of the data.

[0034] For each historical microseismic energy feature subsequence in the historical microseismic energy feature subsequence set, the difference value of the microseismic energy feature value between adjacent two time points is calculated according to the time sequence of the data points in the subsequence. For example, a historical microseismic energy feature subsequence contains energy values of a first time period, energy values of a second time period, energy values of a third time period, and a series of energy data arranged in time sequence. The first-order difference operation process of the subsequence is as follows: subtract the energy value of the first time period from the energy value of the second time period to obtain the first difference value; subtract the energy value of the second time period from the energy value of the third time period to obtain the second difference value; and so on, until all the energy difference values between adjacent time periods are calculated.

[0035] Through the first-order difference operation, the original absolute value information of energy is converted into energy change rate information. Positive value indicates that the microseismic energy in adjacent time period shows an upward trend, negative value indicates that the microseismic energy shows a downward trend, and the size of the difference value reflects the degree of energy change. This difference feature can effectively eliminate the influence of energy benchmark value, highlight the dynamic change characteristics of microseismic energy, and especially can identify abnormal patterns of rapid energy accumulation or sudden release.

[0036] After performing the first-order difference operation on each historical microseismic energy feature subsequence in the historical microseismic energy feature subsequence set, the corresponding historical differential energy feature subsequence is obtained. All these historical differential energy feature subsequences are combined together to form a historical differential energy feature subsequence set. The historical differential energy feature subsequence set and the original historical microseismic energy feature subsequence set complement each other, providing rich feature information for subsequent spatiotemporal orientation analysis and energy accumulation tendency judgment.

[0037] S4, performing trend correlation analysis on the historical microseismic energy feature subsequence set and the historical differential energy feature subsequence set respectively by using a spatiotemporal orientation module to obtain a historical microseismic energy trend feature set and a historical differential energy trend feature set.

[0038] Specifically, the historical microseismic data (the historical microseismic energy feature subsequence set and the historical differential energy feature subsequence set) are subjected to deep trend correlation analysis by the spatiotemporal orientation module, and the change law and correlation characteristics of microseismic energy are mined from the time dimension and the space dimension, so as to obtain a trend feature set (the historical microseismic energy trend feature set and the historical differential energy trend feature set) that can reflect the accumulation tendency of microseismic energy.

[0039] The spatiotemporal orientation module is an intelligent analysis framework based on a feedforward neural network, and includes a time orientation submodule and a space orientation submodule that work in parallel.

[0040] The time orientation submodule is specially used for analyzing the evolution law and trend characteristics of microseismic energy features in time series. The time orientation submodule receives the historical microseismic energy feature subsequence set or the historical differential energy feature subsequence set as input, and identifies the time mode of energy change, such as periodic fluctuation, progressive growth, sudden change, etc., through the internal neural network structure. The time orientation submodule can capture the energy change law at different time scales, including instantaneous change in the short term and cumulative trend in the long term.

[0041] The space orientation submodule is specially used for analyzing the correlation and propagation characteristics of microseismic energy features in spatial distribution. The space orientation submodule combines the hypocenter location information of microseismic events to analyze the energy correlation between different spatial positions and identify the spatial mode of energy accumulation, such as local aggregation, regional diffusion, directional propagation, etc. The space orientation submodule can find the energy interaction and influence mode between adjacent regions.

[0042] In the trend correlation analysis process, the time-oriented sub-module and the space-oriented sub-module work in parallel, respectively extracting features in the time dimension and the space dimension for the same set of input data. For the set of historical microseismic energy feature subsequences, the two sub-modules extract the time trend features and the space trend features respectively, and then integrate the time-space information through a feature fusion mechanism, finally outputting the set of historical microseismic energy trend features. Similarly, for the set of historical differential energy feature subsequences, after the same time-space analysis process, the set of historical differential energy trend features is output. The two trend feature sets comprehensively reflect the complex change law of microseismic energy in the time-space dimension, providing feature input for the subsequent energy accumulation tendency judgment.

[0043] S5, constructing an energy accumulation tendency analyzer based on the set of historical microseismic energy trend features and the set of historical differential energy trend features.

[0044] Specifically, based on the set of historical microseismic energy trend features and the set of historical differential energy trend features, an intelligent analyzer capable of judging the energy accumulation tendency of microseismic energy, i.e., an energy accumulation tendency analyzer, is constructed. The energy accumulation tendency analyzer is a deep learning model based on comparative representation learning and consistency loss calculation, which is used to identify the consistency relationship between microseismic energy features and differential energy features, and to judge whether there is an energy accumulation tendency accordingly.

[0045] The construction process of the energy accumulation tendency analyzer includes steps such as training sample construction, probability distribution mapping, loss function design, and model training. First, the feature data of the corresponding time period in the historical microseismic energy trend feature set and the historical differential energy trend feature set are paired. Each pair of historical microseismic energy trend features and historical differential energy trend features is a set of positive sample pairs, representing the consistency relationship that should be maintained between the two types of features under normal conditions. Through this pairing method, a set of positive sample pairs for model training is constructed. Next, the historical microseismic energy trend feature set and the historical differential energy trend feature set are respectively converted into probability distribution. Through normalization processing and Softmax function mapping operation, the original trend feature value is converted into probability distribution form, so as to eliminate the influence of different feature dimensions and make the subsequent consistency comparison more accurate and stable. Then, a consistency loss function is designed to measure the difference between the two trend features. The consistency loss function is based on the symmetric KL divergence calculation method. By comparing the similarity between the historical microseismic energy trend feature probability distribution and the historical differential energy trend feature probability distribution, the consistency degree of the two is quantified. When the two types of features are highly consistent, the loss value is small; when the two types of features have significant differences, the loss value is large. Then, a deep learning framework based on convolutional neural network is used to optimize the network parameters through supervised learning with the set of positive sample pairs as the training input. During the training process, the network learns to minimize the consistency loss function, so that the model can accurately identify the consistency pattern of microseismic energy features and differential energy features under normal conditions. After training, the model becomes the energy accumulation tendency analyzer. In practical applications, when the consistency loss value calculated from the real-time microseismic data exceeds the preset threshold, the energy accumulation tendency analyzer judges that there is an energy accumulation tendency and outputs a warning signal.

[0046] S6, obtain a set of real-time microseismic log sequences obtained by real-time monitoring of the seismic monitoring sensor array, and perform abnormal identification by combining the energy accumulation tendency analyzer to trigger a microseismic energy accumulation warning instruction.

[0047] Specifically, the energy accumulation tendency analyzer that has been constructed is used to analyze real-time monitoring data online, realizing real-time warning of microseismic energy accumulation.

[0048] The seismic monitoring sensor array continues to maintain a real-time monitoring state and continuously collects microseismic activity information in the target coal mine area. Similar to the historical data collection process, each microseismic event generated during real-time monitoring is recorded as a real-time microseismic log, containing key information such as event occurrence time, magnitude, hypocenter location, and energy release. These real-time microseismic logs are organized in chronological order to form a set of real-time microseismic log sequences. The set of real-time microseismic log sequences reflects the latest state and development trend of microseismic activity in the current period.

[0049] Subsequently, the same feature extraction and processing procedure as the historical data is performed on the obtained real-time microseismic log sequence set. First, energy feature extraction is performed on the real-time microseismic log sequence set to obtain a real-time microseismic energy feature sequence set; then, first-order difference operation is performed on the energy feature sequence set to construct a real-time differential energy feature sequence set; next, the trained spatiotemporal oriented module is used to perform trend correlation analysis on the two types of real-time feature sequences respectively to obtain a real-time microseismic energy trend feature set and a real-time differential energy trend feature set.

[0050] In the anomaly identification stage, the obtained real-time microseismic energy trend feature set and real-time differential energy trend feature set are input into the energy accumulation tendency analyzer for consistency loss value calculation. The energy accumulation tendency analyzer calculates the difference degree between the current real-time features according to the consistency mode learned in the training stage and outputs a consistency loss value set. The consistency loss value set is analyzed, and the loss values greater than or equal to the preset consistency loss value threshold are identified as abnormal situations. When an anomaly is detected, the spatial distribution features of the anomaly are further analyzed, and through the near neighbor region division and abnormal coefficient calculation, it is judged whether there is a regional energy accumulation tendency. Once it is confirmed that there is an energy accumulation tendency, a microseismic energy accumulation warning instruction is triggered immediately, and warning information is sent to the coal mine safety management personnel to take timely corresponding safety protection measures.

[0051] Further, the historical microseismic log set is processed in stages in combination with the log time stamp to obtain a historical microseismic energy feature subsequence set, including:

[0052] S21, energy feature extraction is performed on the historical microseismic log set by the pre-constructed feature extractor to obtain a historical microseismic energy feature set;

[0053] S22, the historical microseismic energy feature set is processed in sequence according to the log time stamp to obtain a historical microseismic energy feature sequence;

[0054] S23, the historical microseismic energy feature sequence is sliced according to a preset sliding window to determine a historical microseismic energy feature subsequence set.

[0055] In a preferred embodiment, first, the historical microseismic log set is subjected to an energy feature extraction operation by a pre-constructed feature extractor. The feature extractor is a data processing module specially used for extracting key energy parameters from original microseismic logs, and is pre-designed and configured according to microseismic physical characteristics and energy accumulation analysis requirements. The feature extractor extracts multi-dimensional energy-related features from each historical microseismic log, mainly including key parameters such as magnitude of the microseismic event, energy release amount, hypocenter coordinates, hypocenter depth, duration, frequency characteristics, etc. Among them, the magnitude reflects the intensity level of the microseismic event; the energy release amount represents the total energy value released by the microseismic event; the hypocenter coordinates identify the three-dimensional spatial position of the microseismic occurrence; and the hypocenter depth describes the occurrence position of the microseismic event in the vertical direction. By performing feature extraction on all logs in the historical microseismic log set, a historical microseismic energy feature set containing rich energy information is obtained.

[0056] Then, the historical microseismic energy feature set is subjected to time series processing according to the log timestamps corresponding to each historical microseismic energy feature. Specifically, all historical microseismic energy features are rearranged in the order of log timestamps to ensure that they are organized in the order of microseismic event occurrence time. During the serialization process, multiple microseismic energy features with the same or similar timestamps are merged and processed to calculate statistical feature values such as average magnitude, total energy release amount, event frequency, etc. Through this serialization process, the originally discrete and distributed historical microseismic energy features are converted into continuous sequences arranged in chronological order, forming a historical microseismic energy feature sequence. This historical microseismic energy feature sequence can clearly reflect the evolution law and trend of microseismic energy over time.

[0057] Subsequently, the historical microseismic energy feature sequence is subjected to slicing processing using a pre-set sliding window to divide the continuous time sequence into multiple sub-sequences with fixed time span. The pre-set sliding window is a time window used to divide the sequence, and its length is set according to the periodic characteristics of microseismic activity and the timeliness requirements of early warning. Typical sliding window lengths are 2 hours, 6 hours, 12 hours or 24 hours. During the sliding window slicing process, starting from the beginning of the historical microseismic energy feature sequence, data segments of equal length are sequentially extracted in chronological order based on the pre-set sliding window length, and each data segment constitutes a historical microseismic energy feature sub-sequence. In order to ensure the continuity and integrity of the data, a certain degree of overlap can be set between adjacent sliding windows, usually 10% to 50% of the sliding window length. Through the sliding window slicing process, the historical microseismic energy feature sequence is decomposed into multiple sub-sequences with consistent length and continuous time, and these sub-sequences constitute a historical microseismic energy feature sub-sequence set, providing a data input format for subsequent difference operation and trend analysis.

[0058] Further, the historical microseismic energy feature sequence is sliced according to a preset sliding window to determine a historical microseismic energy feature subsequence set, including:

[0059] S231, slicing the historical microseismic energy feature sequence according to a preset sliding window to obtain an initial historical microseismic energy feature subsequence set;

[0060] S232, performing adjacent fusion analysis on the initial historical microseismic energy feature subsequence set to obtain the historical microseismic energy feature subsequence set.

[0061] In a preferred embodiment, first, the historical microseismic energy feature sequence is sliced according to a preset sliding window to obtain an initial historical microseismic energy feature subsequence set. Specifically, the time length of the preset sliding window is taken as the slicing unit, and data segments are sequentially intercepted from the starting time point of the historical microseismic energy feature sequence according to a fixed time interval. Each sliding window covers the same time span, ensuring that the sliced subsequences have uniform time length and data structure. The slicing process adopts an equal-interval division method, that is, there is no overlap or gap between adjacent sliding windows, and the entire historical microseismic energy feature sequence is completely divided into several continuous time periods. Through this initial slicing process, an initial historical microseismic energy feature subsequence set is obtained, each subsequence in the set corresponding to a specific time window and containing all the microseismic energy feature data within the time window.

[0062] Then, adjacent fusion analysis is performed on the initial historical microseismic energy feature subsequence set to identify and merge adjacent subsequences with similar features, so as to reduce data redundancy and improve the efficiency of subsequent analysis. In the adjacent fusion analysis process, first, the feature similarity between each two time-adjacent subsequences in the initial historical microseismic energy feature subsequence set is calculated, and the similarity is quantified by comparing the differences between the statistical feature parameters of the two time-adjacent subsequences, such as the average energy value, energy fluctuation variance, peak frequency and the like. When the feature similarity of the two adjacent subsequences exceeds a preset similarity threshold, it is judged that the microseismic activity patterns represented by the two subsequences are basically consistent and meet the fusion condition. The fusion operation is performed by merging the adjacent similar subsequences in the time dimension to form a new subsequence with a longer time span. The new subsequence after fusion contains all the data information of the original two subsequences, and the feature parameters thereof are recalculated by weighted average or cumulative calculation. The entire initial historical microseismic energy feature subsequence set is subjected to traversal adjacent fusion analysis to identify all adjacent subsequence pairs meeting the fusion condition and perform the fusion operation. Through the adjacent fusion processing, the original possible slight fluctuations or noise interference are effectively smoothed, the overall trend feature of the data is enhanced, and the total number of subsequences is reasonably compressed. After the adjacent fusion analysis, an optimized historical microseismic energy feature subsequence set is obtained, which not only maintains the key information of the original data, but also has better data quality and analysis applicability.

[0063] Further, the trend correlation analysis is performed on the historical microseismic energy feature subsequence set and the historical differential energy feature subsequence set by using the space-time oriented module to obtain a historical microseismic energy trend feature set and a historical differential energy trend feature set, including:

[0064] S41, based on a preset time scale and a sample time-oriented training set, a framework based on a feedforward neural network is supervised trained to obtain a time-oriented sub-module;

[0065] S42, based on a preset space scale and a sample space-oriented training set, a framework based on a feedforward neural network is supervised trained to obtain a space-oriented sub-module;

[0066] S43, the time-oriented sub-module and the space-oriented sub-module are connected in parallel to obtain the space-time oriented module.

[0067] In a preferred embodiment, first, a time-oriented sub-module is constructed, which is specifically used to analyze the time evolution law of microseismic energy. Specifically, a preset time scale is first determined, which refers to the standard time granularity for time feature analysis, and is set according to the time periodicity characteristics of microseismic activity. Typical time scales include different time granularities such as hourly, daily, weekly, etc. At the same time, a sample time-oriented training set is constructed according to the determined preset time scale, which contains a large number of labeled microseismic energy time sequence samples and their corresponding time trend labels, which are used to train the neural network to recognize different time variation patterns, such as energy increasing trend, periodic fluctuation, sudden change, etc. The time-oriented sub-module adopts a deep learning framework based on a feedforward neural network, which contains multiple fully connected layers and activation function layers, and can learn complex nonlinear time mapping relationships. In the supervised training process, the microseismic energy time sequence data in the sample time-oriented training set is taken as the input, and the corresponding time trend label is taken as the expected output. The network parameters are continuously adjusted through the back propagation algorithm, so that the network can accurately identify and predict the time evolution trend of microseismic energy. The training process uses the batch gradient descent optimization method, sets appropriate learning rate and training number of rounds, and continues until the network converges and reaches the preset performance indicators. After training is completed, the neural network framework is the time-oriented sub-module, which has the ability to extract time trend features from microseismic energy sequences.

[0068] At the same time, a space-oriented sub-module is constructed, which is used to analyze the spatial distribution law of microseismic energy. Specifically, a preset spatial scale is first determined, which refers to the standard spatial granularity for spatial feature analysis, and is set according to the spatial range of the target coal mine area and the spatial distribution characteristics of microseismic events. Typical spatial scales include different spatial resolutions such as meter level, ten meter level, hundred meter level, etc. According to the determined preset spatial scale, a sample space-oriented training set is constructed, which contains a large number of labeled microseismic event spatial distribution samples and their corresponding spatial trend labels, which are used to train the neural network to recognize different spatial distribution patterns, such as local aggregation, regional diffusion, directional propagation, etc. The space-oriented sub-module also adopts a deep learning framework based on a feedforward neural network, but its network structure is specially optimized for the characteristics of spatial data, and can process multi-dimensional data containing source location coordinates, spatial distance relationships, etc. In the supervised training process, the microseismic event spatial distribution data in the sample space-oriented training set is taken as the input, and the corresponding spatial trend label is taken as the expected output. The network parameters are optimized through a similar back propagation training process. After training is completed, the neural network framework is the space-oriented sub-module, which has the ability to extract spatial trend features from the spatial distribution of microseismic events.

[0069] Then, the time-oriented sub-module and the space-oriented sub-module that have completed training are connected in parallel to form a complete space-time oriented module. The parallel connection means that the two sub-modules work at the same time in the data processing flow, respectively processing the time dimension and the space dimension information of the same set of input data, and each independently extracts the corresponding trend features. Specifically, when the historical microseismic energy feature subsequence set or the historical differential energy feature subsequence set is input into the space-time oriented module, the time-oriented sub-module focuses on analyzing the time evolution characteristics of these sequences, while the space-oriented sub-module focuses on analyzing the spatial distribution characteristics of the corresponding microseismic events. The output results of the two sub-modules are integrated through a feature fusion mechanism to form comprehensive space-time trend features. The feature fusion process adopts a cascading splicing manner to combine the time trend features and the space trend features into a unified feature vector, which contains the change law information of the microseismic energy in both time and space dimensions. Through parallel connection and feature fusion, the space-time oriented module can comprehensively capture the complex space-time evolution pattern of microseismic energy, providing high-quality feature support for accurately identifying energy accumulation tendency.

[0070] Further, the embodiments of the present application also include:

[0071] S44, respectively using the time-oriented sub-module and the space-oriented sub-module to perform space-time feature extraction on the historical microseismic energy feature sequence and the historical differential energy feature sequence, to obtain historical microseismic energy time trend features, historical microseismic energy space trend features, historical differential energy time trend features, and historical differential energy space trend features;

[0072] S45, performing trend correlation analysis on the historical microseismic energy time trend features and the historical microseismic energy space trend features to obtain historical microseismic energy trend features;

[0073] S46, performing trend correlation analysis on the historical differential energy time trend features and the historical differential energy space trend features to obtain historical differential energy trend features.

[0074] In a preferred embodiment, first, the time-oriented sub-module and the space-oriented sub-module that have been constructed are used to respectively perform space-time dimensional feature extraction operations on the historical microseismic energy feature sequence and the historical differential energy feature sequence. Specifically, the time-oriented sub-module receives the historical microseismic energy feature sequence as input, analyzes the evolution law of the sequence in the time dimension through the internal feedforward neural network structure, identifies the time pattern and trend features of energy change, and outputs the historical microseismic energy time trend features. The historical microseismic energy time trend features reflect the change law of the microseismic energy with time, including energy growth rate, fluctuation period, mutation time, and other time-related feature information.

[0075] Meanwhile, the space-oriented sub-module receives the same sequence of historical microseismic energy features as input, but its focus is on analyzing the spatial distribution characteristics of the corresponding microseismic events. By analyzing the spatial information such as the hypocenter coordinates, spatial aggregation degree, distribution directionality, etc. of the microseismic events, the space-oriented sub-module extracts the distribution rules and propagation characteristics of microseismic energy in the spatial dimension, and outputs the historical microseismic energy spatial trend features. The historical microseismic energy spatial trend features reflect the spatial distribution patterns of microseismic energy, including energy aggregation areas, diffusion directions, spatial correlations, and other spatially related feature information.

[0076] Using the same processing flow, the time-oriented sub-module and the space-oriented sub-module respectively perform feature extraction on the sequence of historical differential energy features, and output the historical differential energy time trend features and the historical differential energy spatial trend features. The historical differential energy time trend features reflect the evolution rules of the microseismic energy change rate in the time dimension, while the historical differential energy spatial trend features reflect the distribution characteristics of the energy change rate in the spatial dimension.

[0077] Subsequently, trend correlation analysis is performed on the historical microseismic energy time trend features and the historical microseismic energy spatial trend features extracted from the same data source, with the aim of organically integrating the feature information in the time and spatial dimensions to obtain overall trend features that comprehensively reflect the spatio-temporal evolution rules of microseismic energy. In the trend correlation analysis process, first, the correlation strength between the time trend features and the spatial trend features is calculated to identify the complementarity and consistency of the two types of features in describing the changes in microseismic energy. Then, a feature weight allocation mechanism is used to allocate corresponding fusion weights to the time trend features and the spatial trend features according to their contribution to energy accumulation prediction. Next, the time trend features and the spatial trend features are combined through weighted fusion to form a unified feature vector. This feature vector contains both the temporal evolution information of microseismic energy and its spatial distribution information, and can comprehensively describe the spatio-temporal variation rules of microseismic energy. After the trend correlation analysis process, the historical microseismic energy trend features are obtained, which comprehensively reflect the complex variation patterns of microseismic energy in the spatio-temporal dimension.

[0078] The same trend correlation analysis method as step S45 is used to fuse the historical differential energy time trend feature and the historical differential energy spatial trend feature to obtain the historical differential energy trend feature. Since the differential energy feature reflects the rate of change of microseismic energy, the trend correlation analysis pays more attention to the dynamic characteristics and mutation characteristics of energy change. By calculating the correlation degree between the two kinds of differential trend features, the cooperative mode of the energy change rate in time and space dimensions is identified. After similar feature weight allocation and weighted fusion processing, the historical differential energy trend feature is obtained. This feature comprehensively reflects the spatio-temporal distribution law of the microseismic energy change rate, and can effectively identify the abnormal mode of rapid energy accumulation or sudden release. The historical differential energy trend feature and the historical microseismic energy trend feature complement each other, providing more comprehensive and accurate feature input for subsequent energy accumulation tendency analysis.

[0079] Further, the historical microseismic energy time trend feature and the historical microseismic energy spatial trend feature are subjected to trend correlation analysis to obtain a historical microseismic energy trend feature, including:

[0080] S451, calculating the feature similarity of the historical microseismic energy time trend feature and the historical microseismic energy spatial trend feature to obtain a feature similarity set;

[0081] S452, normalizing the feature similarity set and matrixing the processing result to construct a historical microseismic energy time adjacency matrix;

[0082] S453, convolving the historical microseismic energy time adjacency matrix and the historical microseismic energy spatial trend feature to obtain the historical microseismic energy trend feature.

[0083] In a preferred embodiment, first, the correlation between the historical microseismic energy time trend feature and the historical microseismic energy spatial trend feature is established by calculating the similarity between the two features. Specifically, the historical microseismic energy time trend feature and the historical microseismic energy spatial trend feature are taken as two different dimension feature vectors, and the similarity between them is calculated by inner product operation. Inner product operation can quantify the correlation between two feature vectors in numerical distribution. When the numerical change trends of the two feature vectors are consistent, the inner product value is large, indicating high similarity; when the numerical change trends of the two feature vectors are opposite or irrelevant, the inner product value is small, indicating low similarity. The similarity between the historical microseismic energy time trend feature and the corresponding historical microseismic energy spatial trend feature in all time periods is calculated one by one to obtain the spatio-temporal feature correlation degree in different time periods, which constitutes a feature similarity set. This feature similarity set comprehensively describes the spatio-temporal correlation mode of microseismic energy, and provides a quantitative correlation basis for subsequent feature fusion.

[0084] Subsequently, the feature similarity set is normalized and matrix constructed to obtain a historical microseismic energy time adjacency matrix. The normalization processing refers to scaling all numerical values in the feature similarity set to the standard range of zero to one, eliminating the magnitude difference between different similarity numerical values, and ensuring the numerical stability of subsequent matrix operations. The normalized similarity numerical values are further processed by the Softmax function, so that each similarity value is converted into a probability form between zero and one, and the sum of all similarity values is equal to one. In the matrix construction process, the normalized feature similarity numerical values are organized into a two-dimensional matrix structure according to the time sequence and feature correspondence. The rows and columns of the matrix correspond to different time periods or feature dimensions respectively, and the numerical value of each element in the matrix represents the correlation strength between the time trend feature and the spatial trend feature corresponding to the row and column positions. The constructed historical microseismic energy time adjacency matrix reflects the global correlation relationship between the microseismic energy space-time features, wherein the matrix elements with larger numerical values represent stronger correlation between the corresponding features, and the elements with smaller numerical values represent weaker correlation.

[0085] Subsequently, the convolutional neural network is used to perform convolution operation on the historical microseismic energy time adjacency matrix and the historical microseismic energy spatial trend feature, realizing the deep fusion of space-time features. The convolution operation is a feature fusion method based on convolutional neural network, which can extract local correlation patterns between features by sliding convolution kernel, and combine these local patterns into global feature representation. In the convolution fusion process, the historical microseismic energy time adjacency matrix is used as a weight matrix to guide the weight distribution of feature fusion; the historical microseismic energy spatial trend feature is used as the input feature to provide the spatial information to be enhanced. The convolution operation realizes the weighted feature fusion based on correlation strength by element-wise multiplication and accumulation operation between the adjacency matrix and the spatial trend feature. Specifically, the spatial trend feature corresponding to the elements with larger numerical values in the adjacency matrix will obtain higher weight and occupy a more important position in the fusion result; while the feature corresponding to the elements with smaller numerical values in the adjacency matrix will be weakened accordingly. Through the convolution fusion processing based on the adjacency matrix, the time trend information can be effectively fused into the spatial trend feature to obtain a comprehensive feature representation containing both time evolution law and spatial distribution characteristics. The finally output historical microseismic energy trend feature not only retains the core information of the original spatial trend feature, but also integrates the interaction information from the time trend feature, realizes the effective integration of multi-scale information, and avoids the information redundancy problem caused by simple feature splicing.

[0086] Further, based on the historical microseismic energy trend feature set and the historical differential energy trend feature set, an energy accumulation tendency analyzer is constructed, comprising:

[0087] S51, obtaining a set of positive sample pairs by taking each historical microseismic energy trend feature and historical differential energy trend feature in the set of historical microseismic energy trend features and the set of historical differential energy trend features as a positive sample pair;

[0088] S52, mapping the set of historical microseismic energy trend features and the set of historical differential energy trend features into a historical microseismic energy trend feature probability distribution and a historical differential energy trend feature probability distribution respectively;

[0089] S53, performing symmetric KL divergence calculation on the historical microseismic energy trend feature probability distribution and the historical differential energy trend feature probability distribution to determine a consistency loss function;

[0090] S54, performing supervised calculation on a framework constructed based on a convolutional neural network by taking the set of positive sample pairs as a contrastive representation learning input, and constructing an energy accumulation tendency analyzer by minimizing the consistency loss function.

[0091] In a preferred embodiment, first, a set of positive sample pairs for contrastive representation learning is constructed by pairing combination. Specifically, the feature data corresponding to the same time period in the set of historical microseismic energy trend features and the set of historical differential energy trend features are paired one by one, and each historical microseismic energy trend feature and its corresponding historical differential energy trend feature in the same time period form a positive sample pair. This pairing method is based on the following basis: under normal microseismic activity state, the absolute value feature of microseismic energy and its change rate feature should maintain a certain internal consistency relationship, that is, both should reflect the same microseismic activity pattern and energy evolution trend. By traversing processing, a corresponding positive sample pair is established for each time period, and these positive sample pairs represent the consistency pattern that the microseismic energy trend feature and the differential energy trend feature should maintain under normal state. All positive sample pairs are combined to form a set of positive sample pairs, which provides standard training samples for subsequent contrastive representation learning, so that the energy accumulation tendency analyzer can learn to recognize the consistency relationship pattern between the two features under normal state.

[0092] Subsequently, the historical microseismic energy trend feature set and the historical differential energy trend feature set are respectively converted into corresponding probability distribution forms for subsequent consistency quantization analysis. The probability distribution mapping process includes normalization processing and Softmax function mapping. First, all feature values in the historical microseismic energy trend feature set are normalized to eliminate the influence of dimensional differences on subsequent calculations. Then, the normalized feature values are mapped by the Softmax function to convert each feature value into a probability value between zero and one, and ensure that the sum of all probability values in the same feature vector is equal to one. The same processing procedure is used to map the probability distribution of the historical differential energy trend feature set, obtaining the historical differential energy trend feature probability distribution. Through this probability distribution mapping process, the original trend feature data is converted into a standardized probability distribution form, providing a unified data format for subsequent consistency loss calculation, and making the comparison between different types of features more fair and accurate.

[0093] Subsequently, a consistency loss function is designed based on the symmetric KL divergence calculation method to quantify the difference between the historical microseismic energy trend feature probability distribution and the historical differential energy trend feature probability distribution. Symmetric KL divergence is a mathematical index that measures the similarity between two probability distributions, and its value reflects the difference between the two distributions: when the two probability distributions are completely consistent, the symmetric KL divergence value is zero; when the two probability distributions are significantly different, the symmetric KL divergence value is larger. By calculating the symmetric KL divergence between the microseismic energy trend feature probability distribution and the differential energy trend feature probability distribution in each positive sample pair, the loss value reflecting the consistency of the two features is obtained. This loss value is used as the output of the consistency loss function to guide the subsequent model training process. The design goal of the consistency loss function is to maintain a high degree of consistency between the microseismic energy trend feature and the differential energy trend feature under normal conditions, and the corresponding loss value should be small; while in the abnormal state, the consistency between the two features is destroyed, and the corresponding loss value should increase significantly.

[0094] After that, the contrastive representation learning method is used to train the deep learning framework based on the convolutional neural network, and the energy accumulation tendency analyzer is constructed by minimizing the consistency loss function. Contrastive representation learning is a machine learning method that learns data representation by comparing the similarities and differences between samples, which is suitable for learning the consistency relationship pattern between data. During the training process, the set of positive sample pairs is used as input data, and the microseismic energy trend features and the differential energy trend features in each positive sample pair are input into the convolutional neural network framework for feature coding and representation learning. Through multi-layer convolution operation and nonlinear transformation, the network learns to extract deep representations of the two features and calculates the consistency loss value between the two features. The goal of training is to minimize the consistency loss function value of the positive sample pair by adjusting the network parameters, so that the network can accurately identify the consistency pattern between the two features in the normal state. After sufficient supervised training, the deep learning framework becomes the energy accumulation tendency analyzer. In the actual deployment and application stage, the energy accumulation tendency analyzer receives the real-time microseismic energy trend features and differential energy trend features to be detected as input, and calculates the consistency loss value between the two. When the calculated consistency loss value exceeds the preset threshold value, it is judged that the current microseismic activity state deviates from the normal consistency pattern, there is an energy accumulation tendency, and the corresponding early warning discrimination result is output. Otherwise, when the consistency loss value is within the normal range, the analyzer judges that the current state is normal and there is no significant energy accumulation risk.

[0095] Further, the set of historical microseismic energy trend features and the set of historical differential energy trend features are respectively mapped into a historical microseismic energy trend feature probability distribution and a historical differential energy trend feature probability distribution, comprising:

[0096] S521, normalizing and softmax mapping the set of historical microseismic energy trend features to obtain a historical microseismic energy trend feature probability distribution;

[0097] S522, normalizing and softmax mapping the set of historical differential energy trend features to obtain a historical differential energy trend feature probability distribution.

[0098] In a preferred embodiment, firstly, the historical microseismic energy trend feature set is subjected to probability distribution conversion processing. Specifically, firstly, all feature data in the historical microseismic energy trend feature set is subjected to normalization processing, the maximum value and the minimum value of all numerical values in the historical microseismic energy trend feature set are calculated, then the maximum and minimum value normalization method is adopted, each feature numerical value is linearly scaled according to its relative position in the overall numerical value range, so that all feature numerical values are mapped into the standard interval of zero to one. Through normalization processing, the numerical magnitude difference between different feature dimensions can be eliminated, and the numerical stability of subsequent processing can be ensured. After normalization processing, the processing result is subjected to softmax mapping operation. Softmax mapping is a mathematical transformation method for converting an arbitrary real number vector into a probability distribution, the basic principle of which is to perform exponential function transformation on each element in the vector, then divide the transformation result by the sum of all element exponential transformation results, so as to obtain a probability distribution with a sum of one. Through softmax mapping, the normalized historical microseismic energy trend feature is converted into a probability form, in which each feature numerical value corresponds to a probability value between zero and one, and the sum of all probability values in the same feature vector is strictly equal to one. After normalization and softmax mapping processing, the historical microseismic energy trend feature probability distribution is obtained, which maintains the relative relationship of the original feature data and has the standard mathematical properties of the probability distribution.

[0099] Meanwhile, the same processing procedure as step S521 is adopted to perform probability distribution conversion on the historical differential energy trend feature set. First, normalization processing is performed on all feature data in the historical differential energy trend feature set, the maximum value and the minimum value of the historical differential energy trend feature set are calculated, and then the same maximum and minimum value normalization method is adopted to linearly scale all numerical values of the differential energy trend feature to the standard range of zero to one. Next, the normalized differential energy trend feature data is subjected to softmax mapping transformation, and each feature vector is converted into a corresponding probability distribution form. Since the differential energy trend feature reflects the change rate information of microseismic energy, its numerical value may contain positive and negative values, and the normalization processing can effectively process such bidirectional change numerical characteristics, and the softmax mapping ensures that the final probability distribution has standard mathematical properties. After the same normalization and softmax mapping processing, the historical differential energy trend feature probability distribution is obtained. The historical differential energy trend feature probability distribution and the historical microseismic energy trend feature probability distribution have the same mathematical format and properties, both of which are expressed in a standardized probability form, providing format-unified and numerically stable input data for subsequent symmetric KL divergence calculation and consistency analysis. Through symmetric probability distribution mapping processing, the absolute value feature and the change rate feature of the microseismic energy are converted into a probability distribution form that can be directly compared, so that the two different features can be accurately quantified and analyzed for consistency.

[0100] Further, as shown in Figure 2 The real-time microseismic log sequence set obtained by real-time monitoring of the seismic monitoring sensor array is acquired, and the energy accumulation tendency analyzer is combined to identify abnormalities and trigger a microseismic energy accumulation warning instruction, including:

[0101] S61, feature extraction is performed on the real-time microseismic log sequence set to determine a real-time microseismic energy feature sequence set;

[0102] S62, first-order differential operation is performed on the real-time microseismic energy feature sequence set to determine a real-time differential energy feature sequence set;

[0103] S63, the real-time microseismic energy feature sequence set and the real-time differential energy feature sequence set are subjected to trend correlation analysis by the space-time oriented module to determine a real-time microseismic energy trend feature set and a real-time differential energy trend feature set;

[0104] S64, the real-time microseismic energy trend feature set and the real-time differential energy trend feature set are subjected to consistency loss value analysis by the energy accumulation tendency analyzer to determine a consistency loss value set;

[0105] S65, analyze the consistency loss value set, if there is an anomaly, trigger microseismic energy accumulation warning instruction.

[0106] In a preferred embodiment, first, the real-time microseismic log sequence set obtained by real-time monitoring of the seismic monitoring sensor array is processed by feature extraction, and a real-time microseismic energy feature sequence set is determined. The processing flow of this step is the same as the feature extraction process of historical data, but the processing object is the real-time collected microseismic monitoring data. Specifically, the same feature extractor as step S21 is used to extract energy-related parameters for each microseismic log in the real-time microseismic log sequence set, including magnitude, energy release, focal position coordinates, focal depth, duration, and other key feature information of microseismic events. Then, according to the same sequence processing method as step S22, the extracted energy features are arranged in chronological order according to the timestamp information of the real-time microseismic log, forming a real-time microseismic energy feature sequence. Then, using the same sliding window slicing processing method as step S23, the real-time microseismic energy feature sequence is segmented according to the preset sliding window length, and adjacent fusion analysis is performed, and finally a real-time microseismic energy feature sequence set is obtained. This real-time microseismic energy feature sequence set reflects the real-time change state of microseismic energy in the current period, providing a data basis for subsequent real-time analysis.

[0107] Subsequently, the same first-order difference operation method as step S3 is used to process the real-time microseismic energy feature sequence set. For each feature sequence in the real-time microseismic energy feature sequence set, the difference between the energy feature values of the adjacent two time points is calculated in chronological order, and the difference information reflecting the current microseismic energy change rate is obtained. Through first-order difference operation, the dynamic change trend of microseismic energy can be captured in real time, especially the abnormal change pattern of rapid energy accumulation or sudden release can be identified in time. After processing, a real-time difference energy feature sequence set is obtained, which complements the real-time microseismic energy feature sequence set and provides more comprehensive real-time microseismic activity feature information.

[0108] Then, the real-time microseismic energy feature sequence set and the real-time differential energy feature sequence set are respectively subjected to trend correlation analysis by using the spatiotemporal orientation module which has been trained. The processing method of this step is the same as that of step S4, but the processing object is real-time data. The time orientation submodule and the space orientation submodule in the spatiotemporal orientation module work in parallel, respectively performing deep analysis on real-time feature data from the time dimension and the space dimension, and extracting the spatiotemporal evolution law and trend features of the current microseismic activity. Through the same feature extraction and fusion process as steps S44 to S46, real-time microseismic energy time trend features, real-time microseismic energy space trend features, real-time differential energy time trend features and real-time differential energy space trend features are obtained, and then through trend correlation analysis, a real-time microseismic energy trend feature set and a real-time differential energy trend feature set are finally obtained. The two trend feature sets comprehensively reflect the complex change law of the current microseismic activity in the spatiotemporal dimension.

[0109] Subsequently, the real-time microseismic energy trend feature set and the real-time differential energy trend feature set are subjected to consistency loss value analysis by using the energy accumulation tendency analyzer which has been constructed. The energy accumulation tendency analyzer receives the two types of real-time trend features as input, and calculates the consistency degree between the current real-time microseismic energy trend features and the real-time differential energy trend features according to the consistency judgment mode learned in the training phase. Specifically, the real-time trend features are converted into probability distribution form, and then the symmetric KL divergence between the two probability distributions is calculated to obtain a loss value reflecting the consistency degree of the current state. The same consistency loss value calculation is performed on the real-time trend features of all time periods to obtain a consistency loss value set. Each loss value in the consistency loss value set corresponds to a specific time period and a spatial region, and reflects the consistency degree of the microseismic energy features and the differential features in the spatiotemporal range. When the consistency loss value is small, it indicates that the current state conforms to the normal microseismic activity mode; when the consistency loss value is large, it indicates that the current state deviates from the normal mode, and there may be an energy accumulation tendency.

[0110] Then, the consistency loss value set is analyzed to determine whether there is an abnormal situation and to determine whether to trigger a warning instruction. First, each loss value in the consistency loss value set is compared with a preset consistency loss value threshold to identify all abnormal loss values greater than or equal to the threshold. Then, the monitoring location and time information corresponding to the abnormal loss values are analyzed to determine the spatial distribution characteristics and time development trend of the abnormalities. When abnormal loss values are detected and the spatial aggregation degree and duration of the abnormal loss values both reach the preset warning conditions, it is determined that the current microseismic activity has a significant energy accumulation tendency, and a microseismic energy accumulation warning instruction is immediately triggered. The warning instruction includes key information such as the time, location, and severity of the abnormal detection, and sends warning information to the coal mine safety management personnel through various ways such as audible and visual alarms, SMS notifications, and system interface prompts, so as to take appropriate safety protection and emergency response measures in a timely manner.

[0111] Further, the energy accumulation tendency analyzer is used to analyze the consistency loss values of the real-time microseismic energy trend feature set and the real-time differential energy trend feature set to determine a consistency loss value set, including:

[0112] S641, obtaining the monitoring coordinates corresponding to the consistency loss values greater than or equal to the preset consistency loss value threshold in the consistency loss value set to obtain an abnormal monitoring coordinate set;

[0113] S642, dividing the abnormal monitoring coordinate set into K partitioned neighbor regions according to a preset monitoring coordinate neighbor threshold, wherein K is a positive integer;

[0114] S643, traversing and counting the number of abnormal monitoring coordinates in the K partitioned neighbor regions, and obtaining K partitioned neighbor region abnormal coefficients by placing the statistical results in the area of the K partitioned neighbor regions;

[0115] S644, when any one of the K partitioned neighbor region abnormal coefficients is greater than or equal to a preset partitioned neighbor region abnormal coefficient threshold, a microseismic energy accumulation warning instruction is triggered.

[0116] In a preferred embodiment, firstly, the spatial position information with abnormality is identified from the consistency loss value set. Specifically, firstly, all loss values in the consistency loss value set are obtained, and each loss value is compared with a preset consistency loss value threshold one by one. The preset consistency loss value threshold is a critical value determined according to the statistical analysis result of historical normal microseismic activity data. When the real-time calculated consistency loss value exceeds the threshold, it indicates that the consistency between the microseismic energy features and the difference features of the current space-time position is destroyed, and there is an abnormal energy accumulation tendency. For each abnormal loss value greater than or equal to the preset consistency loss value threshold, the monitoring coordinate information corresponding to the abnormal loss value is extracted, including the three-dimensional spatial coordinates of the microseismic event occurrence position associated with the loss value. These monitoring coordinates identify the specific spatial position of abnormal microseismic activity in the target coal mine area. By traversing the entire consistency loss value set, all abnormal loss values corresponding to the monitoring coordinates are collected to form an abnormal monitoring coordinate set. The abnormal monitoring coordinate set provides basic data for subsequent spatial clustering analysis, so that the distribution characteristics and concentration degree of abnormality can be judged from the spatial dimension.

[0117] Then, the abnormal monitoring coordinate set is subjected to spatial clustering analysis based on a preset monitoring coordinate near neighbor threshold, the purpose of which is to merge the abnormal monitoring coordinates with similar spatial positions into the same region to lay a foundation for subsequent accumulation risk analysis. The preset monitoring coordinate near neighbor threshold refers to a spatial distance critical value for judging whether two monitoring coordinates belong to a near neighbor relationship. The threshold is set according to the geological structure characteristics of the target coal mine area and the layout density of the microseismic monitoring sensor, and a typical value is 50 meters to 200 meters. A distance-based clustering algorithm is used to divide the near neighbor regions of the abnormal monitoring coordinate set. Specifically, the three-dimensional spatial distance between any two coordinate points in the abnormal monitoring coordinate set is calculated, and when the distance between the two coordinate points is less than or equal to the preset monitoring coordinate near neighbor threshold, it is judged that the two coordinate points belong to the same near neighbor region. Through this distance-constrained clustering method, all abnormal monitoring coordinates are grouped according to spatial proximity, and each group constitutes a divided near neighbor region. After the near neighbor region division processing, K divided near neighbor regions are obtained, where K is a positive integer, and the specific value depends on the spatial distribution characteristics of the abnormal monitoring coordinates and the size of the preset near neighbor threshold. Each divided near neighbor region contains several abnormal monitoring coordinates with similar spatial positions. This region division method can effectively identify the spatial clustering pattern of abnormality and provide an important basis for judging the regional characteristics of energy accumulation.

[0118] Subsequently, the concentration of each partitioned near-neighbor region is calculated by statistical analysis to obtain a quantitative regional anomaly coefficient. First, traverse the K partitioned near-neighbor regions, and count the number of anomaly monitoring coordinates contained in each region, which reflects the concentration of abnormal events in the region. Then, calculate the coverage area of each partitioned near-neighbor region, i.e. the area size of the spatial range formed by all anomaly monitoring coordinates in the region. Next, divide the number of anomaly monitoring coordinates in each partitioned near-neighbor region by the coverage area of the region to obtain the partitioned near-neighbor region anomaly coefficient of the region. The partitioned near-neighbor region anomaly coefficient is a dimensionless value reflecting the anomaly density per unit area, and its numerical size directly reflects the concentration of abnormal events in the region. When the partitioned near-neighbor region anomaly coefficient of a region is high, it indicates that there are more abnormal events per unit area in the region, and there is significant spatial aggregation, which may indicate a strong energy accumulation tendency in the region. By performing the same anomaly coefficient calculation on the K partitioned near-neighbor regions, K partitioned near-neighbor region anomaly coefficients are obtained, providing a quantitative basis for the final warning judgment.

[0119] Subsequently, the final warning decision is made based on the regional anomaly coefficient to determine whether to trigger the microseismic energy accumulation warning instruction. Specifically, compare the K partitioned near-neighbor region anomaly coefficients with the preset partitioned near-neighbor region anomaly coefficient threshold value respectively. The preset partitioned near-neighbor region anomaly coefficient threshold value is a critical value determined according to statistical analysis of historical microseismic disaster events and expert experience. When the regional anomaly coefficient exceeds this threshold value, it indicates that the concentration of anomalies in the region has reached a dangerous level that may trigger energy accumulation risk. When it is found that any one of the K partitioned near-neighbor region anomaly coefficients is greater than or equal to the preset partitioned near-neighbor region anomaly coefficient threshold value, the microseismic energy accumulation warning instruction is triggered immediately. This judgment mechanism is based on the concept of regional risk assessment, i.e. even if there is high-density abnormal aggregation in a local region, it is enough to constitute a whole security threat, and a warning needs to be issued in time. The triggering of the warning instruction contains key information such as the specific location information of the abnormal region, the anomaly coefficient value, the warning level, etc., which provides accurate risk positioning and severity assessment for coal mine safety management personnel, and facilitates the adoption of targeted safety protection measures.

[0120] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0121] Those skilled in the art will appreciate that embodiments of the present application can be devised for a variety of applications. It is intended that the present application be limited only by the scope of the appended claims, and it is intended that various modifications and alterations made by those skilled in the art be considered as within the scope of the present application. The embodiments of the present application will be described with reference to the attached drawings identified below.

[0122] The present application is described in reference to the drawings using a flowchart and / or a block diagram of the method, apparatus (system) and computer program product according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagram, and combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0123] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0124] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0125] Although preferred embodiments of the application have been described, those skilled in the art will recognize that additional modifications and variations are possible without departing from the spirit and scope of the application.

[0126] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the application, the application can be practiced otherwise than as specifically described.

Claims

1. A microseismic energy accumulation tendency-based early warning method, characterized in that, The method comprises: acquiring a set of historical microseismic logs collected by a seismic monitoring sensor array arranged in a target coal mine region within a historical time window, wherein each historical microseismic log comprises a log timestamp; processing the set of historical microseismic logs in stages in combination with the log timestamps to obtain a set of historical microseismic energy feature subsequences; performing first-order difference operations on the set of historical microseismic energy feature subsequences respectively to construct a set of historical differential energy feature subsequences; performing trend correlation analysis on the set of historical microseismic energy feature subsequences and the set of historical differential energy feature subsequences respectively using a space-time guiding module to obtain a set of historical microseismic energy trend features and a set of historical differential energy trend features; constructing an energy accumulation tendency analyzer based on the set of historical microseismic energy trend features and the set of historical differential energy trend features; comprising: obtaining a set of positive sample pairs by taking each historical microseismic energy trend feature and historical differential energy trend feature in the set of historical microseismic energy trend features and the set of historical differential energy trend features as a positive sample pair; mapping the set of historical microseismic energy trend features and the set of historical differential energy trend features into a historical microseismic energy trend feature probability distribution and a historical differential energy trend feature probability distribution respectively; performing symmetric KL divergence calculation on the historical microseismic energy trend feature probability distribution and the historical differential energy trend feature probability distribution to determine a consistency loss function; performing supervised calculation on a framework constructed based on a convolutional neural network by taking the set of positive sample pairs as a comparative representation learning input, constructing an energy accumulation tendency analyzer by minimizing the consistency loss function; obtaining a set of real-time microseismic log sequences obtained by real-time monitoring of the seismic monitoring sensor array, and performing abnormality identification in combination with the energy accumulation tendency analyzer to trigger a microseismic energy accumulation early warning instruction.

2. The microseismic energy accumulation tendency-based early warning method according to claim 1, characterized in that, processing the set of historical microseismic logs in stages in combination with the log timestamps to obtain a set of historical microseismic energy feature subsequences, comprising: performing energy feature extraction on the set of historical microseismic logs using a pre-constructed feature extractor to obtain a set of historical microseismic energy features; performing serialization processing on the set of historical microseismic energy features according to the log timestamps to obtain a historical microseismic energy feature sequence; performing slicing on the historical microseismic energy feature sequence according to a preset sliding window to determine the set of historical microseismic energy feature subsequences.

3. The microseismic energy accumulation tendency-based early warning method according to claim 2, characterized in that, performing slicing on the historical microseismic energy feature sequence according to a preset sliding window to determine the set of historical microseismic energy feature subsequences, comprising: performing slicing on the historical microseismic energy feature sequence according to a preset sliding window to obtain an initial set of historical microseismic energy feature subsequences; performing adjacent fusion analysis on the initial set of historical microseismic energy feature subsequences to obtain the set of historical microseismic energy feature subsequences.

4. The microseismic energy accumulation tendency-based early warning method according to claim 1, characterized in that, performing trend correlation analysis on the set of historical microseismic energy feature subsequences and the set of historical differential energy feature subsequences respectively using a space-time guiding module to obtain a set of historical microseismic energy trend features and a set of historical differential energy trend features, comprising: Supervised training is performed on the framework constructed based on the feedforward neural network based on a preset time scale and a sample time-oriented training set, to obtain a time-oriented sub-module; Supervised training is performed on the framework constructed based on the feedforward neural network based on a preset space scale and a sample space-oriented training set, to obtain a space-oriented sub-module; The time-oriented sub-module and the space-oriented sub-module are connected in parallel to obtain the spatio-temporal orientation module.

5. The microseismic energy accumulation tendency-based early warning method according to claim 4, characterized in that, It comprises: The time-oriented sub-module and the space-oriented sub-module are connected in parallel to obtain the spatio-temporal orientation module. The historical microseismic energy time trend feature and the historical microseismic energy space trend feature are subjected to trend correlation analysis to obtain a historical microseismic energy trend feature. The historical microseismic energy time trend feature and the historical microseismic energy space trend feature are subjected to trend correlation analysis to obtain a historical microseismic energy trend feature.

6. The microseismic energy accumulation tendency-based early warning method according to claim 5, characterized in that, The historical microseismic energy time trend feature and the historical microseismic energy space trend feature are subjected to trend correlation analysis to obtain a historical microseismic energy trend feature. The historical microseismic energy time trend feature and the historical microseismic energy space trend feature are subjected to trend correlation analysis to obtain a historical microseismic energy trend feature, comprising: The feature similarity of the historical microseismic energy time trend feature and the historical microseismic energy space trend feature is calculated to obtain a feature similarity set; The feature similarity set is subjected to normalization processing, and the processing result is matrixed to construct a historical microseismic energy time adjacency matrix; 7. The microseismic energy accumulation tendency-based early warning method according to claim 6, characterized in that, The historical microseismic energy time adjacency matrix and the historical microseismic energy space trend feature are convolved to obtain the historical microseismic energy trend feature. The historical microseismic energy trend feature set and the historical differential energy trend feature set are respectively mapped into a historical microseismic energy trend feature probability distribution and a historical differential energy trend feature probability distribution, comprising: The historical microseismic energy trend feature set is subjected to normalization and softmax mapping to obtain a historical microseismic energy trend feature probability distribution; 8. The microseismic energy accumulation tendency-based early warning method according to claim 1, characterized in that, The historical differential energy trend feature set is subjected to normalization and softmax mapping to obtain a historical differential energy trend feature probability distribution. The real-time microseismic log sequence set obtained by real-time monitoring of the seismic monitoring sensor array is combined with the energy accumulation tendency analyzer to identify an anomaly and trigger a microseismic energy accumulation warning instruction, comprising: Energy feature extraction is performed on the real-time microseismic log sequence set to determine a real-time microseismic energy feature sequence set; First-order differential operation is performed on the real-time microseismic energy feature sequence set to determine a real-time differential energy feature sequence set; Trend correlation analysis is performed on the real-time microseismic energy feature sequence set and the real-time differential energy feature sequence set using the spatio-temporal orientation module to determine a real-time microseismic energy trend feature set and a real-time differential energy trend feature set; Consistency loss value analysis is performed on the real-time microseismic energy trend feature set and the real-time differential energy trend feature set using the energy accumulation tendency analyzer to determine a consistency loss value set; The consistency loss value set is analyzed, and if there is an anomaly, a microseismic energy accumulation warning instruction is triggered.

9. The microseismic energy accumulation tendency-based early warning method according to claim 8, characterized in that, The real-time microseismic energy trend feature set and the real-time differential energy trend feature set are analyzed by the energy accumulation tendency analyzer to determine a consistency loss value set, including: A monitoring coordinate corresponding to a consistency loss value greater than or equal to a preset consistency loss value threshold in the consistency loss value set is obtained to obtain an abnormal monitoring coordinate set; The abnormal monitoring coordinate set is divided into K division neighbor regions according to a preset monitoring coordinate neighbor threshold, wherein K is a positive integer; The number of abnormal monitoring coordinates in the K division neighbor regions is counted, and the statistical result is the area of the K division neighbor regions to obtain K division neighbor region abnormal coefficients; When any one of the K division neighbor region abnormal coefficients is greater than or equal to a preset division neighbor region abnormal coefficient threshold, a microseismic energy accumulation warning instruction is triggered.

Images