CN-121596370-B - Mine microseism focus positioning method, system, equipment and storage medium based on multiple intelligent agents

CN121596370BCN 121596370 BCN121596370 BCN 121596370BCN-121596370-B

Abstract

The invention discloses a mine microseism focus positioning method, system, equipment and storage medium based on multiple agents, wherein the method comprises the steps of configuring a corresponding sub-agent for each detector in a mine area, inputting initial focus prediction coordinates into a pre-constructed virtual microseism positioning environment, starting an iterative loop to enable each agent to perform positioning search in the virtual microseism positioning environment, responding to the fact that the current focus prediction coordinates output by the sub-agents after current iteration meet preset iteration stopping conditions, outputting the current focus prediction coordinates as positioning results of the sub-agents, conducting aggregation based on the positioning results of the sub-agents, outputting target focus coordinates of microseism events, realizing modeling a microseism focus positioning process as a multi-agent collaborative optimization process, realizing iteration and optimization of focus coordinates through reinforcement learning strategies, realizing dynamic adaptation to the mine complex environment, and remarkably improving the precision, stability and intelligent level of mine microseism positioning.

Inventors

CHEN XIAOHONG
CAI CHENGCHENG
AN QINGXIAN
ZHANG YUHANG
LUO WENJIE
Xie Zeqiang

Assignees

中南大学

Dates

Publication Date: 20260508
Application Date: 20260130

Claims (9)

1. The mine microseismic source positioning method based on the multiple agents is characterized by comprising the following steps of: Configuring a corresponding sub-intelligent agent for each detector in a mine area, and constructing an intelligent agent model corresponding to each sub-intelligent agent based on signal data of the detector, wherein the sub-intelligent agent comprises three space dimension coordinate intelligent agents; generating a priori constraint range of a seismic source space based on waveform image data of the microseismic signals received by each detector, and initial seismic source prediction coordinates of each sub-agent; inputting the initial seismic source prediction coordinates into a pre-constructed virtual microseism positioning environment, generating initial states of all coordinate agents through the virtual microseism positioning environment, and starting an iterative loop to enable all coordinate agents to output respective coordinate increment in the virtual microseism positioning environment based on the current states and historical messages of other agents, wherein the virtual microseism positioning environment is constructed based on a multi-agent reinforcement learning environment, and the searching range of the coordinate agents in the virtual microseism positioning environment is the prior constraint range; Responding to the fact that current focus prediction coordinates output by sub-agents after current iteration meet preset iteration stop conditions, outputting the current focus prediction coordinates as positioning results of the sub-agents, wherein the current focus prediction coordinates are determined based on coordinate increment of the coordinate agent output of three space dimensions of the sub-agents; aggregating based on the positioning results of the plurality of sub-agents, and outputting the target focus coordinates of the microseism event; The generating the prior constraint range of the source space and the initial source prediction coordinates of each sub-agent based on the waveform image data of the microseismic signals received by each detector comprises the following steps: Acquiring detector signal data of a mine area, wherein the detector signal data comprises corresponding spatial position coordinates of a detector buried in the mine and waveform image data of microseismic signals received by the detector; carrying out band-pass filtering and self-adaptive denoising on the waveform image data acquired by each detector to obtain waveform characteristics; Inputting the characteristics of the waveform to a pre-trained mixed deep learning model, and outputting first arrival moments of longitudinal waves and transverse waves, wherein the mixed deep learning model is constructed based on a convolutional neural network and a long-term and short-term memory network; determining the wave travel time difference of each detector based on the first arrival time of the longitudinal wave and the transverse wave; And determining the interval distance between the microseismic event and each detector based on the three-dimensional speed model of the mine area and the wave travel time difference, and generating a priori constraint range of a seismic source space and an initial seismic source prediction coordinate based on the interval distance.
2. The multi-agent based mine microseismic source location method of claim 1 wherein prior to inputting the initial source prediction coordinates into a pre-constructed virtual microseismic location environment, further comprising: Constructing a virtual microseismic positioning environment based on the prior constraint range and the sub-intelligent bodies corresponding to the detectors; The input data of the virtual microseismic positioning environment comprises spatial position coordinates of each detector, first arrival time of longitudinal waves and transverse waves, a three-dimensional speed model of a mine area and a focus prediction coordinate, when the virtual microseismic positioning environment is configured to calculate a microseismic theory through the three-dimensional speed model and an approximate straight line propagation model, global state vectors for each coordinate intelligent agent to call in the virtual microseismic positioning environment comprise residual statistics features, weighted residual features, geometric distribution features, historical actions and convergence trends, a total reward function configured by the virtual microseismic positioning environment comprises basic error items, a focus physical reachable area penalty item, a focus depth priori reward item and action smoothness constraint items, the total reward function is used for evaluating action effects of the coordinate intelligent agents so as to perform reward and punishment on search actions of the coordinate intelligent agents, the virtual microseismic positioning environment is further configured with a collaborative mechanism used for configuring a strategy network and a value network for each coordinate intelligent agent, and the virtual microseismic positioning environment is further configured with a message mechanism used for enabling each coordinate intelligent agent in the three spatial dimensions in the same sub intelligent agent to achieve coordinate adjustment.
3. The multi-agent-based mine microseismic source positioning method according to claim 2, wherein the mathematical expression corresponding to the basic error term is: Wherein, the Is the first The error type prize value for the next iteration, And For non-negative weight parameters, controlling the influence weights of the longitudinal wave residual error and the transverse wave residual error on the total rewards respectively, And The sum of squares of the residual errors of the longitudinal waves and the sum of the residual errors of the transverse waves of all detectors are respectively calculated; the mathematical expression corresponding to the source physical reachable region penalty term is as follows: Wherein, the Is the first The source physical reachable region penalty term for the next iteration, Is the source physical reachable region penalty term coefficient, Is an indication function of the display, Is the first The source prediction coordinates of the multiple iterations, Is the effective area physically accessible to the seismic source; The mathematical expression corresponding to the seismic source depth priori rewarding item is as follows: Wherein, the Is the first The source depth of the next iteration is a priori awarded items, Is a priori weighting coefficient, for controlling the effect of a priori terms in the total prize, Is the current predicted depth The corresponding a priori probability density is used, Is a natural logarithm of the number of the pairs, Is the first Z-direction coordinates of source predictions at a number of iterations; The mathematical expression corresponding to the action smoothness constraint term is as follows: Wherein, the Is the first The action smoothness of the next iteration rewards the term, Is the motion difference penalty coefficient, Is the first The coordinate correction amount of the number of iterations, Is the first The coordinate correction amount of the number of iterations, The square of the L2 norm is used for representing the vector length of the difference between the two correction amounts; the mathematical expression corresponding to the total rewarding function is as follows: Wherein, the Represent the first Total prize value for each iteration.
4. The mine microseismic source positioning method based on multiple agents of claim 3, further comprising, after the constructing a virtual microseismic positioning environment based on the prior constraint range and the sub-agents corresponding to each detector: Randomly generating a plurality of seismic source positions in a simulation environment based on a three-dimensional speed model of a mine area and a known roadway or goaf structure, and randomly sampling the seismic source three-dimensional coordinates of the seismic source positions generated by simulation in a physical reachable area inside an actual mine; Calculating theoretical observation time of three-dimensional coordinates of each seismic source, superposing random noise and pickup error on the theoretical observation time, and obtaining simulation observation time of longitudinal waves and transverse waves, wherein the following formula is referred to: Wherein, the And Respectively the first in the simulation environment When the simulation of the longitudinal wave and the transverse wave of each detector is observed, And Respectively with three-dimensional coordinates of seismic source For input, according to theoretical arrival time of longitudinal wave and transverse wave obtained by calculating three-dimensional speed model, And Random noise and pickup errors added to longitudinal waves and transverse waves are used for simulating measurement errors in a real environment, and the noise is simulated according to Gaussian distribution; Constructing an input sample based on simulation observation of the longitudinal wave and the transverse wave, inputting the input sample into a virtual microseismic positioning environment, and performing reinforcement learning training on each agent in the virtual microseismic positioning environment based on the total rewarding function; Generating a semi-supervised fine tuning dataset based on historical microseismic event samples, the historical microseismic event samples comprising observed times of historical microseismic events and true source location results; And inputting the semi-supervised fine tuning data set into the virtual microseismic positioning environment to update the step length of the agent after reinforcement learning training.
5. The multi-agent-based mine microseismic source positioning method of any one of claims 1 to 4, wherein after inputting the initial source prediction coordinates into a pre-constructed virtual microseismic positioning environment, generating initial states of each coordinate agent through the virtual microseismic positioning environment, and starting an iterative loop, further comprising: Judging whether the current focus prediction coordinates output by the sub-agent after the current iteration meet preset iteration stop conditions or not, wherein the preset iteration stop conditions comprise residual norms calculated based on the current focus prediction coordinates are smaller than a preset convergence threshold; if the residual norm is not smaller than the preset convergence threshold, judging that the current seismic source prediction coordinates do not meet the preset iteration stop condition, and continuing the iteration cycle; And if the residual norm is smaller than the preset convergence threshold, judging that the current focus prediction coordinate meets the preset iteration stop condition, stopping the iteration loop, executing the step of responding to the fact that the current focus prediction coordinate output by the sub-agent after the current iteration meets the preset iteration stop condition, and outputting the current focus prediction coordinate as a positioning result of the sub-agent.
6. The multi-agent based mine microseismic source location method of any one of claims 1 to 4 wherein the aggregating based on the location results of the plurality of sub-agents outputs the target source coordinates of the microseismic event comprising: The weight of the sub-agent corresponding to each detector is calculated based on the signal-to-noise ratio of each detector, and the following formula is referred to: Wherein, the Is a child intelligent agent The signal-to-noise ratio of the signal, The number of sub-agents is indicated, Representing sub-agents Weights of (2); aggregating the positioning results of the plurality of sub-agents based on the weights of the sub-agents, outputting the target source coordinates of the microseismic event, and referring to the following formula: Wherein, the Is the target source coordinates of the microseismic event, Is the positioning result of the microseismic event prediction output after the iteration of the ith sub-agent is completed, Representing the number of iterations.
7. Mine microseismic source positioning system based on many agents, its characterized in that, mine microseismic source positioning system based on many agents includes: The intelligent agent configuration module is used for configuring a corresponding sub intelligent agent for each detector in the mine area, and constructing an intelligent agent model corresponding to each sub intelligent agent based on the signal data of the detector, wherein the sub intelligent agent comprises three space dimension coordinate intelligent agents; the data analysis module is used for generating a priori constraint range of a seismic source space and initial seismic source prediction coordinates of each sub-agent based on waveform image data of the micro-seismic signals received by each detector; The positioning reasoning module is used for inputting the initial seismic source prediction coordinates into a pre-constructed virtual microseismic positioning environment, generating initial states of all coordinate agents through the virtual microseismic positioning environment, starting an iterative loop so that all coordinate agents output respective coordinate increment in the virtual microseismic positioning environment based on the current states and historical messages of other agents, the virtual microseismic positioning environment is constructed based on a multi-agent reinforcement learning environment, and the searching range of the coordinate agents in the virtual microseismic positioning environment is the prior constraint range; the positioning output module is used for outputting the current focus prediction coordinates output by the sub-agents after the current iteration as the positioning result of the sub-agents in response to the fact that the current focus prediction coordinates output by the sub-agents meet the preset iteration stop condition, and the current focus prediction coordinates are determined based on the coordinate increment of the coordinate agent output of the three space dimensions of the sub-agents; the seismic source positioning module is used for carrying out aggregation based on positioning results of a plurality of sub-agents and outputting target seismic source coordinates of a microseismic event; The data analysis module is further used for acquiring detector signal data of a mine area, the detector signal data comprise corresponding space position coordinates of detectors buried in the mine and waveform image data of microseismic signals received by the detectors, the waveform image data acquired by the detectors are subjected to band-pass filtering and self-adaptive denoising processing to obtain waveforms, the waveforms are characterized, the waveforms are input into a pre-trained mixed deep learning model, first arrival moments of longitudinal waves and transverse waves are output, the mixed deep learning model is constructed based on a convolutional neural network and a long-short term memory network, wave travel time differences of the detectors are determined based on the first arrival moments of the longitudinal waves and the transverse waves, the distance between a microseismic event and the detectors is determined based on a three-dimensional speed model of the mine area and the wave travel time differences, and a priori constraint range of a seismic source space and initial seismic source prediction coordinates are generated based on the distance.
8. A multi-agent-based mine microseismic source positioning device, comprising a memory, a processor and a multi-agent-based mine microseismic source positioning program stored on the memory, wherein the processor is used for running the multi-agent-based mine microseismic source positioning program, and the multi-agent-based mine microseismic source positioning program is configured to implement the multi-agent-based mine microseismic source positioning method as set forth in any one of claims 1 to 6.
9. A computer-readable storage medium, wherein the computer-readable storage medium has stored thereon a multi-agent-based mine microseismic source positioning program, which when executed by a processor, implements the multi-agent-based mine microseismic source positioning method according to any one of claims 1 to 6.

Description

Mine microseism focus positioning method, system, equipment and storage medium based on multiple intelligent agents Technical Field The invention relates to the technical field of mine safety, in particular to a mine microseism focus positioning method, system, equipment and storage medium based on multiple intelligent agents. Background The mine microseism monitoring technology is a key link in the current mine safety production, and disaster events such as rock burst, roof caving, rock burst and the like can be effectively early warned through real-time monitoring of rock mass breakage, stress adjustment and dynamic disaster precursor activities in the mine. The vibration source positioning precision is one of the most core technical indexes of the micro-vibration monitoring system, and the precision directly influences judgment of disaster occurrence parts, range and danger level. The existing mine microseismic positioning method is generally based on the steps that microseismic waveform signals are collected through an underground detector array, P, S wave first arrival time is picked up, and a three-dimensional coordinate of a seismic source is solved through an iterative inversion or search algorithm in combination with a mine three-dimensional speed model. However, the underground mine environment is complex, the speed structure is non-uniform and continuously changes along with mining activities, and the detector signals are interfered by strong industrial noise, fan noise and mechanical vibration, so that the traditional positioning method based on analytical models or linearization inversion has obvious limitations in practical engineering. For example, the internal speed structure of a mine is continuously changed along with goaf generation, surrounding rock damage and geologic structure change, so that calculation errors are large in theoretical running, stable accuracy is difficult to ensure in a traditional method, a speed model is difficult to construct accurately, a gradient descent or least square inversion-based method is strong in dependence on initial seismic source guess, when initial value deviation is large, a local optimal solution is easy to fall into, historical positioning experience cannot be effectively fed back to a positioning algorithm along with the change of the mine structure, a positioning strategy is difficult to optimize continuously along with time, and a cooperative and weighting mechanism is absent among detectors. The existing mine microseismic positioning technology mainly comprises a traditional three-dimensional seismic source inversion method, but the following defects still exist in practical application: the traditional positioning method is highly dependent on the accuracy of a mine three-dimensional speed model, when the model has local deviation or a speed structure is complex (fault, goaf and dense roadway), the travel time residual error is difficult to converge, and the positioning error is obviously increased. The mine environment has a large amount of mechanical noise and background disturbance, so that the pickup error of a low signal-to-noise ratio microseismic event is amplified, the traditional method lacks robustness, and particularly when the signal quality of part of detectors is poor, the whole positioning result is easy to be dragged and deviated. The traditional positioning only depends on a single model to directly output three-dimensional coordinates, the coupling relationship between the three dimensions is weak, so that an error in a certain direction can drive the whole offset, and an information feedback mechanism between coordinate dimensions is lacked. Disclosure of Invention The invention mainly aims to provide a mine microseismic source positioning method, system, equipment and storage medium based on multiple intelligent agents, and aims to solve the technical problems that in the prior art, microseismic positioning is low in precision, poor in robustness and convergence under the conditions of complex speed structure, high noise, incomplete data and insufficient three-dimensional coordinate coupling, and cannot be dynamically adapted to complex environments of mines, so that the mine microseismic positioning precision is low and stability is poor. In order to achieve the purpose, the invention provides a mine microseism focus positioning method based on multiple intelligent agents, which comprises the following steps: Configuring a corresponding sub-intelligent agent for each detector in a mine area, and constructing an intelligent agent model corresponding to each sub-intelligent agent based on signal data of the detector, wherein the sub-intelligent agent comprises three space dimension coordinate intelligent agents; generating a priori constraint range of a seismic source space based on waveform image data of the microseismic signals received by each detector, and initial seismic source prediction coordinates of each sub-agent;