A multi-index fusion discrimination mine earthquake adaptive positioning method and system

Abstract

A multi-index fusion discriminant mine earthquake adaptive positioning method and system belong to the field of mine earthquake positioning, the system includes data acquisition module, feature extraction and quality evaluation module, comprehensive index calculation module and adaptive decision module. The system first carries out multidimensional physical examination on the input waveform data, station layout and velocity model, generates a quantitative "physical examination report" containing signal-to-noise ratio, pickup error, station geometry coverage, model reliability and the like, that is, a multi-dimensional feature vector; then, a built-in rule engine will automatically select the optimal single method or combined strategy under the current conditions from a variety of positioning methods according to the report and the user's set requirement target, and output the decision result with clear basis. The present application realizes the full-process automation from data input to strategy output through the above-mentioned system and method, and provides efficient and reliable technical support for mine safety monitoring and disaster warning.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of mine earthquake positioning, and relates to a mine earthquake adaptive positioning method and system based on multi-index fusion discrimination. BACKGROUND

[0002] Mine earthquake (or microseismic) monitoring is one of the key technologies for preventing mine dynamic disasters, ensuring the safety of underground operation personnel and the stable operation of mine production facilities. One of its core tasks is to achieve rapid and accurate positioning of the earthquake source. The effective selection and fusion of mine earthquake positioning methods as an important technical means to improve positioning reliability and adapt to complex mine environments are directly related to the timeliness of disaster warning, the accuracy of positioning results and the effectiveness of emergency response. Therefore, building an intelligent positioning system that can dynamically adapt to data conditions and fuse the advantages of multiple methods is a key direction and urgent need to break through the current positioning bottleneck and improve mine safety monitoring capabilities.

[0003] At present, the commonly used mine earthquake positioning methods have certain limitations in complex actual environments, and most of them can only achieve good results under specific conditions or ideal assumptions, and it is difficult to maintain high precision and high robustness in varying working conditions. Specifically, the travel time positioning method has high calculation efficiency, but its positioning effect is severely dependent on accurate phase picking under high signal-to-noise ratio conditions, good station geometric layout and single-component sensor data. In actual noisy interference, uneven station coverage or picking deviation, positioning divergence or significant error increase is easy to occur; the full waveform migration stacking method has high tolerance to picking errors and certain robustness when the signal-to-noise ratio is low and the number of stations is limited, but its positioning accuracy is significantly affected by the accuracy of the velocity model, and the calculation burden increases with the complexity of the model; the Bayesian probability method can provide complete posterior probability distribution and uncertainty quantification information, and the results are more reliable, but the calculation cost is extremely high, and the data quality and station number requirements are strict, making it difficult to meet real-time monitoring needs. In addition, existing positioning systems generally lack comprehensive perception and dynamic evaluation of multi-source information (such as signal-to-noise ratio level, picking error, station spatial distribution, velocity model reliability and calculation resource constraints), and fail to achieve adaptive method selection and multi-strategy collaborative positioning based on actual observation conditions, resulting in insufficient stability of the system in extreme or non-ideal working conditions, long-tailed error distribution, and serious constraints on the real-time, accuracy and overall reliability of the positioning task in mine safety monitoring.

[0004] The accurate positioning of mine earthquake is a key technical link of mine dynamic disaster monitoring and early warning, and the accuracy and reliability of the result directly affect the effectiveness of disaster identification and emergency decision. However, the traditional positioning method often performs well only under specific data conditions or ideal assumptions, and is difficult to cope with the fluctuating signal-to-noise ratio, pickup error, limited station distribution and uncertain velocity model in the actual mine environment, resulting in large positioning error and even failure in extreme scenarios. Although the deep learning positioning method has excellent effect, it has poor generalization and needs to be retrained in different mining areas, which reduces the practicability. The existing technology lacks comprehensive perception of multi-source heterogeneous information and fails to establish a dynamic decision-making and multi-method coordination mechanism based on real-time working conditions, which seriously restricts the accuracy and stability of the mine safety monitoring system in practical application. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a multi-index fusion discrimination mine earthquake adaptive positioning method and system, which adaptively selects the optimal positioning strategy and performs accurate analysis through event waveform data, station location information and velocity model.

[0006] To solve the above technical problems, the technical content of the present application is: a multi-index fusion discrimination mine earthquake adaptive positioning system, comprising a data acquisition module, a feature extraction and quality evaluation module, a comprehensive index calculation module and an adaptive decision module.

[0007] The data acquisition module is used for receiving and preprocessing the seismic event waveform data of multiple stations, station spatial coordinate information and velocity model parameter information, and outputting the waveform data and metadata in a unified format to the feature extraction and quality evaluation module. Then the P-wave arrival time is obtained through the automatic phase picking algorithm of the module, and the signal-to-noise ratio and pickup uncertainty of each station are calculated, and the results are output to the comprehensive index calculation module. The comprehensive index calculation module receives the primary parameters from the feature extraction and quality evaluation module, performs comprehensive calculation, generates a multi-dimensional feature vector including the number of effective stations, average signal-to-noise ratio, average pickup uncertainty, station gap angle, velocity model reliability, data type identification and resource availability, and outputs the feature vector to the adaptive decision module. Through the rule engine and positioning method library built-in the adaptive module, combined with the multi-dimensional feature vector received from the comprehensive index calculation module, according to the preset rules and decision mode, the optimal positioning strategy is dynamically selected or combined, and the final positioning scheme is output.

[0008] A method for multi-index fusion discrimination mine earthquake adaptive positioning by using a multi-index fusion discrimination mine earthquake adaptive positioning system, comprising the following steps:

[0009] Step 1, system initialization and data preprocessing: the system receives multi-station seismic waveform data and pre-processes the data. Based on the pre-processed waveform data, the system uses classical signal detection and picking algorithms to automatically pick the P-wave first arrival of each station waveform, and synchronously calculates the signal-to-noise ratio SNR and picking uncertainty σt of each station.

[0010] The specific method is:

[0011] Step 1.1, the system receives multi-station seismic waveform data supporting single-component and three-component formats, station information files containing spatial coordinates and component type metadata, velocity model parameters including model type, grid spacing and calibration residual, and grid parameters including spatial resolution;

[0012] Step 1.2, determine whether the waveform recorded by each station is single-component or three-component; for three-component data, the system preferentially uses the Z component for subsequent processing, and if there is no Z component, it generates an equivalent single-component waveform by EN energy stacking;

[0013] Step 1.3, based on the pre-processed waveform data, use classical signal detection and picking algorithms to automatically pick the P-wave first arrival of each station waveform;

[0014] The STA / LTA method is based on the significant change in energy characteristics of seismic signals before and after the arrival of P waves. It realizes the first arrival detection by calculating the ratio of average energy in the sliding time window, and its formula is expressed as:

[0015]

[0016]

[0017] wherein is the waveform amplitude, and are the short-time window and long-time window lengths respectively, and the AIC method segments the waveform by using a local autoregressive model to find the point that minimizes the information as the phase arrival time, and its expression is

[0018]

[0019] Calculate the short-time window / long-time window of the waveform signal, i.e. the STA / LTA amplitude average ratio. When the ratio exceeds the preset threshold, it is determined as a possible P-wave first arrival, and the preliminary arrival time is selected. In the time window near the first arrival triggered by STA / LTA, the Akaike Information Criterion AIC is used to further analyze the signal to determine the optimal position of the waveform feature point, and the P-wave arrival time is picked up. The ratio method trigger combined with the AIC refinement algorithm for automatic P-wave arrival time picking, synchronously calculates the signal-to-noise ratio SNR and picking uncertainty σt of each station.

[0020] Step 2, Key parameter extraction and quality assessment: Based on the data results in step 1, calculate the key indicators for evaluating the overall quality and applicability of the current data set, including: number of effective stations, signal-to-noise ratio, pick uncertainty, maximum station gap angle, and velocity model reliability.

[0021] The specific method is:

[0022] Step 2.1, signal-to-noise ratio calculation and effective station number

[0023] The signal-to-noise ratio is defined as the logarithmic form of the ratio of the signal energy in the P-wave window to the energy in the noise window:

[0024]

[0025] Where: add parameter meaning;

[0026] The pick uncertainty is estimated based on the curvature of the AIC function near the arrival time or the standard deviation of multiple pick results,

[0027] Define the station with signal-to-noise ratio SNR≥5.0 and σt≤0.04s as an effective station, only select the signal effective station to participate in positioning, the average signal-to-noise ratio is the average of the SNR values of all effective stations, the average pick uncertainty σt of the P-wave arrival time is the average value of the P-wave arrival time pick uncertainty σt of all effective stations, reflecting the reliability of the overall P-wave arrival time pick, the smaller σt, the more reliable the timing basis of positioning.

[0028] Step 2.2, station gap angle analysis and heat map construction

[0029] The station gap angle AG represents the maximum angle blank area formed by the station direction vector in the horizontal plane projection at any point in the monitoring area when observed, i.e. the maximum uncovered angle between adjacent station azimuth angles, reflecting the geometric coverage of the station to the source point in the plane: the smaller the gap angle, the more uniform the station coverage, the higher the positioning accuracy; on the contrary, if the gap angle is large, there is a significant monitoring blind area, which is easy to cause the source positioning error to increase or miss detection;

[0030] The specific calculation process is as follows:

[0031] For any source point in the monitoring area , first calculate the direction vector of each station :

[0032]

[0033] Then, convert each direction vector to azimuth angle :

[0034]

[0035] Sort the azimuths of all stations in ascending order, and calculate the included angle between the adjacent two azimuths:

[0036]

[0037] The included angle between the last station and the first station is:

[0038]

[0039] Finally, take the maximum value of all included angles as the maximum gap angle at the source point:

[0040]

[0041] By calculating and statistics the AG value of potential source location in the monitoring area of three-component station envelope, a two-dimensional AG heat map is generated to show the uniformity of station direction coverage at different locations in the monitoring area;

[0042] Define that when the average gap angle ≤90°, the spatial distribution of the station is better, and the monitoring coverage is more sufficient; if the regional gap angle >270°, it indicates that there is a significant lack of monitoring direction at this place, by adding stations or adjusting the layout of existing stations to eliminate the monitoring blind area.

[0043] Step 2.3, velocity model credibility analysis

[0044] The credibility of the velocity model is used to quantify the priori index of the characterization ability of the current velocity model to the real underground wave velocity structure. This index is based on the comprehensive evaluation of the complexity and spatial resolution of the model. The core principle is: under the premise that the model is effectively constrained, the higher the complexity and the finer the spatial resolution of the model, the stronger the ability to depict the actual geological heterogeneity, and thus it is given a higher credibility;

[0045] The system uses the following determined and quantitative evaluation criteria to assign credibility to the input velocity model:

[0046] Step 2.3.1. Mean velocity model

[0047] The entire monitoring area is simplified as a homogeneous medium, containing only one uniform velocity value; the complexity of this model is the lowest, and its credibility is fixed and quantified as 0.3;

[0048] Step 2.3.2. One-dimensional layered velocity model

[0049] Assuming that the velocity model only varies with depth, it can depict the layered sedimentary sequence of the stratum, and the credibility of the model is linearly quantified according to the number of its layers: when the number of layers ≥ 5, the credibility V_conf = 0.7; when 2 ≤ the number of layers < 5, the credibility V_conf = 0.4 + 0.1 × (the number of layers - 2); when the number of layers < 2, it is classified as an average model, and the credibility is 0.3;

[0050] Step 2.3.3. Three-dimensional structural velocity model

[0051] The model can represent the anisotropy and non-uniformity distribution of velocity in three-dimensional space; the credibility of the model is determinedly quantified according to its grid spacing: if the grid spacing > 50 m, then V_conf = 0.7; if 30 m < grid spacing ≤ 50 m, then V_conf = 0.8; if 15 m < grid spacing ≤ 30 m, then V_conf = 0.9; if 5 m < grid spacing ≤ 15 m, then V_conf = 0.95; if the grid spacing ≤ 5 m, then V_conf = 1;

[0052] The velocity model credibility V_conf is an independent quantitative index, which is input into the rule engine of the adaptive decision module together with the "data type identification, i.e. single / three-component", "effective station number N_eff" and other dimensional features;

[0053] The rule engine will comprehensively consider all these parameters and evaluate the applicability of different positioning methods according to the pre-set and determined rules.

[0054] Step 3, rule engine and strategy decision: the system is built-in with the travel time inversion positioning method Powell, the waveform migration stacking positioning method QM, and the Bayesian positioning method MaxEnt_HMC. Based on the indicators in step 2, the system makes dynamic decisions through the rule engine, and outputs a single method or a cascading strategy;

[0055] Minimum time delay mode: first, if the input data is three-component data, the system directly disables the Powell method; if the Powell method is not disabled and its admission condition is met, the single Powell positioning method is preferentially selected; otherwise, if the admission condition of the QM method is met, the single QM positioning method is then selected; otherwise, if the admission condition of the MaxEnt_HMC method is met, the single MaxEnt_HMC positioning method is selected;

[0056] Precision-first mode: if both Powell and MaxEnt_HMC methods are allowed to be cascaded and their admission conditions are satisfied, the system recommends and executes Powell → MaxEnt_HMC cascaded positioning strategy; otherwise, if both QM and MaxEnt_HMC methods are allowed to be cascaded and their admission conditions are satisfied, the system recommends and executes QM → MaxEnt_HMC cascaded positioning strategy; otherwise, if the admission condition of MaxEnt_HMC method is satisfied, the system recommends and executes single MaxEnt_HMC positioning method as fallback;

[0057] Need-uncertainty mode: the system aims to ensure the output of complete posterior distribution and uncertainty measure as the first priority, and the decision logic is as follows: if both Powell and MaxEnt_HMC methods are allowed to be cascaded and their admission conditions are satisfied, the system recommends and executes Powell → MaxEnt_HMC cascaded strategy, which first uses the Powell method to quickly obtain a high-precision initial source parameter, and then uses the MaxEnt_HMC method to sample in the neighborhood of this initial solution to efficiently obtain refined posterior distribution and uncertainty; otherwise, if both QM and MaxEnt_HMC methods are allowed to be cascaded and their admission conditions are satisfied, the system recommends and executes QM → MaxEnt_HMC cascaded strategy, which first uses the QM method to determine a high-probability source spatial coherence region, and then uses the MaxEnt_HMC method to sample in this coherence region to robustly obtain posterior distribution and uncertainty; otherwise, if the admission condition of MaxEnt_HMC method is satisfied, the system directly recommends and executes single MaxEnt_HMC positioning method to output posterior distribution and uncertainty;

[0058] Comprehensive mode: in this mode, the system aims to balance the positioning speed, accuracy and uncertainty information of the results, and makes dynamic decisions according to the user's demand for posterior distribution, and the logic is as follows:

[0059] a. When the user does not require the output of posterior distribution and uncertainty, if the admission condition of Powell method is satisfied, the system recommends and executes single Powell positioning method to achieve fast positioning; otherwise, if the admission condition of QM method is satisfied, the system recommends and executes single QM positioning method; otherwise, if the admission condition of MaxEnt_HMC method is satisfied, the system recommends and executes single MaxEnt_HMC positioning method;

[0060] b.When the user requires the output of the posterior distribution and the uncertainty, if the cascade permission is true, and the access conditions of the QM method and the MaxEnt HMC method are both met, the system preferentially recommends and executes the QM to MaxEnt HMC cascade positioning strategy, which firstly determines the high-probability source coherent area by using the QM method, and then samples in the area by using the MaxEnt HMC method, so as to balance the efficiency and the reliability of the uncertainty quantification, otherwise, if the access condition of the MaxEnt HMC method is met, the system recommends and executes the single MaxEnt HMC positioning method.

[0061] Step 4, after the system decision is completed, the interpretation output and the report generation link are entered, the decision logic in the system, the key index evaluation results and the finally recommended positioning strategy are output in a structured and readable manner, and a decision report is generated.

[0062] The application has the advantages that: the application takes multi-source information fusion and dynamic optimization as the core, firstly integrates multi-dimensional indexes such as data quality, station geometry, model reliability, data type and computing resources into a unified decision framework, and constructs a systematic and quantitative evaluation and selection mechanism. The method breaks through the limitation of the poor adaptability of the traditional single positioning method in a complex environment, realizes the change from the 'fixed strategy' to the 'condition-driven', and significantly improves the comprehensive performance of the positioning system.

[0063] Specifically, the application establishes an accurate quantitative standard for the applicability of the positioning method by real-time sensing of key parameters such as signal-to-noise ratio level, pickup uncertainty, effective station number, azimuth gap angle and velocity model reliability, and combining three-component data automatic identification and resource state monitoring. The system can dynamically adjust the decision strategy according to the actual data conditions, such as automatically switching from the travel time inversion method with high data quality requirements to the waveform migration superposition method with stronger fault tolerance when the station signal is lost or the signal-to-noise ratio drops sharply, thereby effectively enhancing the robustness and stability of the system under non-ideal working conditions.

[0064] Especially prominent is that the application clearly proposes the applicable conditions of different positioning methods, ensuring the theoretical rigor and result reliability of the positioning process. At the same time, the Chinese explainable report generated by the system elaborates the decision basis and index values in detail, greatly enhancing the transparency and user trust of the system.

[0065] In addition, the application provides a plurality of configurable strategy modes (such as minimum time delay, precision priority, uncertainty requirement, etc.), which can flexibly adapt to the differentiated requirements of different application scenarios from real-time emergency early warning to scientific research precise inversion, and realize the optimal balance between calculation efficiency, positioning accuracy and result reliability. The system realizes the full-process automation from data input to strategy output, significantly reduces the dependence on professional experience, and provides efficient and reliable technical support for mine safety monitoring and disaster warning. BRIEF DESCRIPTION OF DRAWINGS

[0066] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings. Among them:

[0067] Figure 1 Adaptive decision flowchart for earthquake positioning method

[0068] Figure 2 Input continuous waveform data example diagram

[0069] Figure 3 Mine area station distribution plan

[0070] Figure 4 Layered uniform velocity model diagram of mine area

[0071] Figure 5 P-wave first arrival picking diagram

[0072] Figure 6 Arrival time picking uncertainty diagram

[0073] Figure 7 Station gap angle thermal diagram

[0074] Figure 8 Powell inversion positioning principle diagram

[0075] Figure 9 Schematic diagram of migration stacking method

[0076] Figure 10 Hamilton Monte Carlo positioning principle diagram DETAILED DESCRIPTION

[0077] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0078] Embodiment 1

[0079] I. Adaptive decision system based on feature quantization and rule reasoning

[0080] The decision system integrates multiple positioning algorithms: nonlinear optimization algorithm Powell based on travel time information, offset stacking algorithm QM based on full waveform information, and Bayesian inference method MaxEnt HMC based on probability statistics. The system adaptively selects the optimal positioning strategy and performs accurate analysis through event waveform data, station location information and velocity model.

[0081] The system first discriminates the format of the input data to distinguish whether it is full single-component waveform data. Then, in the feature extraction and quality evaluation module, the STA / LTA and AIC joint automatic picking algorithm is used to obtain the P-wave arrival time, and the signal-to-noise ratio, picking uncertainty and other quality parameters of each station are calculated synchronously. On this basis, the comprehensive index calculation module generates a multi-dimensional feature vector including the number of effective stations, average signal-to-noise ratio, average picking uncertainty, azimuth gap angle, velocity model reliability, data type identification and resource availability.

[0082] In the adaptive decision module, the system makes a comprehensive judgment on the above indicators through the rule engine. For coal mine working face microseismic monitoring, the commonly used travel time positioning methods include Geiger algorithm and Powell algorithm. Due to the limited monitoring range, small change in wave velocity structure, the travel time picking error is usually not large, and the positioning accuracy can meet the standard requirements. However, when introducing three-component stations and expanding the monitoring coverage, the travel time picking error and the uncertainty of the propagation path increase significantly, resulting in a decrease in positioning accuracy. Therefore, the system automatically disables the travel time positioning algorithm in the non-single component data scenario to control the overall error level. After excluding the inapplicable methods, the system then performs preliminary screening according to the admission threshold of each positioning method. For example, the QM method requires effective station number ≥3 and average signal-to-noise ratio ≥4.0 dB, velocity model reliability ≥0.5; the Powell method requires effective station number ≥4 and average signal-to-noise ratio ≥6.0 dB, velocity model reliability ≥0.5, maximum station gap angle ≤180°, and picking uncertainty <0.04 s; the MaxEnt_HMC method requires effective station number ≥4 and average signal-to-noise ratio ≥6.0, velocity model reliability ≥0.5, and picking uncertainty <0.04 s. Subsequently, combined with the user-set strategy mode (such as minimum time delay, precision priority, uncertainty constraint, or comprehensive balance), the system completes the final strategy recommendation and outputs a single method or cascaded positioning scheme. Finally, the result and report generation module outputs the structured decision result and Chinese explanation, clearly presenting the recommended basis and key indicators, significantly improving the transparency and operability of the system. Through multi-dimensional indicator perception, rule-based logical reasoning, and strategy-based output scheduling, the system realizes intelligent optimization and adaptive switching of positioning methods, effectively enhancing the reliability and practical value of the mine seismic positioning system in complex industrial environments. The system workflow is shown in Figure 1

[0083] II. The decision-making method based on the above system is specifically used to generate an adaptive decision report. The system first performs multi-dimensional "physical examination" on the input waveform data, station layout, and velocity model, generating a quantitative "physical examination report" (i.e., multi-dimensional feature vector) containing signal-to-noise ratio, picking error, station geometric coverage, model reliability, etc. Then, an internal "rule brain" (rule engine) will automatically select the optimal single method or combined strategy under the current conditions from multiple positioning methods based on the report and user-set demand targets (such as fastest, most accurate, or requiring uncertainty), and output the decision result with clear basis.

[0084] The method mainly includes the following steps:

[0085] 1. System initialization and data preprocessing

[0086] ​Firstly, the system receives seismic waveform data of multiple stations (supporting single-component and three-component formats), station information files (containing spatial coordinates, component types, etc. metadata), velocity model parameters (including model type, grid spacing, and calibration residuals), and grid parameters (spatial resolution).

[0087] Firstly, it is determined whether the waveform recorded by each station is single-component or three-component. For three-component data, the system prioritizes the Z component for subsequent processing. If there is no Z component, an equivalent single-component waveform is generated by EN energy stacking to ensure data format uniformity and compatibility. Taking a blast data from Hongqinghe Coal Mine as an example, the mine has 8 monitoring stations, all of which are three-component stations. Among them, 7 stations can normally monitor the waveform, and the normally recorded waveform is shown in Figure 2 , the station distribution and velocity model are shown in Figures 3-4 .

[0088] Based on the preprocessed waveform data, the system uses classical signal detection and picking algorithms to automatically pick the P-wave first arrival of each station waveform.

[0089] The STA / LTA method is based on the significant change in energy characteristics of seismic signals before and after the arrival of P-wave. It realizes the initial arrival detection by calculating the ratio of average energy in the sliding time window. Its formula can be expressed as:

[0090]

[0091] where is the waveform amplitude, and are the short-time window and long-time window length respectively. The AIC method segments the waveform through a local autoregressive model, finds the point that minimizes the information as the phase arrival time, and its expression

[0092] is:

[0093]

[0094] In recent years, relevant scholars' research shows that the STA / LTA and AIC joint picking strategy can effectively combine the advantages of both in noise resistance and segmentation accuracy, significantly improving the robustness and accuracy of P-wave picking. Therefore, this study adopts this combined strategy for phase identification.

[0095] STA / LTA (short-term / long-term average) amplitude ratio of the waveform signal is calculated, and when the ratio exceeds a preset threshold, it is determined that the P-wave first arrival has arrived, and a preliminary arrival time is selected. In the time window near the first arrival triggered by STA / LTA, the AIC (Akaike Information Criterion) criterion is further used to analyze the signal in detail to determine the optimal position of the waveform feature point, so as to realize high-precision picking of the P-wave arrival time. The automatic P-wave arrival time picking by the ratio method combined with AIC refinement algorithm is shown in FIG. 1. Figure 5 and Figure 6 The signal-to-noise ratio SNR and the picking uncertainty σt of each station are calculated synchronously, and details are shown in Table 1.

[0096] Table 1 Summary of average picking uncertainty and station effectiveness

[0097]

[0098] From Table 1, it can be seen that the clear waveform is successfully recorded by 7 three-component stations in the explosion event, the overall signal-to-noise ratio is high, and the average signal-to-noise ratio reaches 15.74, indicating that the waveform quality is good. In terms of picking accuracy, only the picking uncertainty of station D2 is relatively large, and the rest of the stations show high accuracy. According to the defined effective station determination standard, except for D2 and D4, the rest of the stations meet the requirements, so the number of effective stations is 5 (Neff = 5).

[0099] 2. Key parameter extraction and quality evaluation

[0100] Based on the P-wave picking results of the previous step, station information and velocity model, multiple key indicators for evaluating the overall quality and applicability of the current data set are calculated:

[0101] (1) Signal-to-noise ratio calculation and effective station number

[0102] The signal-to-noise ratio is defined as the logarithmic form of the ratio of the signal energy in the P-wave window to the energy in the noise window:

[0103]

[0104] The picking uncertainty is estimated based on the curvature of the AIC function near the arrival time or the standard deviation of multiple picking results.

[0105] The station with a signal-to-noise ratio SNR≥5.0 and σt<0.04s is considered as an effective station. The number of effective stations reflects the number of "high-quality" station resources available for positioning, and is one of the key factors in determining whether certain high-precision methods can be used. Only stations with high signal-to-noise ratio are selected for positioning to improve the reliability of the positioning result.

[0106] The average signal-to-noise ratio is the average of the SNR values of all valid stations (i.e. SNR ≥ 5). The average signal-to-noise ratio is a measure of the average level of overall station signal quality, and the higher the SNR, the better the positioning accuracy is generally.

[0107] The average pick-up uncertainty is the average of the P-wave arrival time pick-up uncertainty σt of all valid stations. It reflects the reliability of the overall P-wave arrival time pick-up, and the smaller σt, the more reliable the timing basis for positioning is.

[0108] (2) Station gap angle analysis and heat map construction

[0109] The station gap angle (AG) is an important indicator for measuring the uniformity of the station planar distribution. It represents the maximum angle blank area formed by the station direction vector in the horizontal plane of any point in the monitoring area when observed from that point, i.e. the maximum uncovered angle between adjacent station azimuths. This angle reflects the geometric coverage of the station to the source point in the plane: the smaller the gap angle, the more uniform the station coverage is, and the higher the positioning accuracy is generally; on the contrary, if the gap angle is large, it means that there is a significant monitoring blind area, which may lead to increased source positioning error or missed detection.

[0110] The specific calculation process is as follows:

[0111] For any source point in the monitoring area , first calculate the direction vector of each station :

[0112]

[0113] Then, convert each direction vector to azimuth :

[0114]

[0115] Sort the azimuths of all stations in ascending order, and calculate the included angle between the adjacent two azimuths:

[0116]

[0117] The included angle between the last station and the first station is:

[0118]

[0119] Finally, take the maximum value of all included angles as the maximum gap angle at the source point:

[0120]

[0121] The AG values of potential source locations in the monitoring area of the 7 three-component station envelopes of this blasting event are calculated and statistically analyzed to generate a two-dimensional AG heat map, as shown in FIG. 3, which intuitively shows the station direction coverage uniformity at different locations in the monitoring area. The warmer the color, the greater the gap angle and the weaker the station geometric coverage, and there is a large monitoring direction blind area. The cooler the color, the smaller the gap angle, the more uniform the station direction distribution, and the more continuous the monitoring coverage. Figure 7

[0122] Experience shows that when the average gap angle is ≤90°, the station spatial distribution uniformity is good, and the monitoring coverage is sufficient. If the gap angle in some areas is >270°, it indicates that there is a significant monitoring direction missing in that area, and the station should be added or the existing station layout should be adjusted to eliminate the monitoring blind area and improve the overall monitoring efficiency and positioning reliability. The average station gap angle of the 7 stations is 134.7°, and the station distribution is not uniform, and the local area may have insufficient monitoring capacity.

[0123] (3) Velocity model credibility analysis

[0124] The system root user pre-inputs the velocity model used for credibility analysis, and the velocity model credibility is an independent index for adaptive decision-making. The velocity model credibility is a key priori evaluation index in this adaptive decision-making framework, which is used to quantify the representation ability of the current velocity model for the true underground wave velocity structure. The comprehensive evaluation of this index is mainly based on the complexity and spatial resolution of the model. The basic theoretical basis is that under the premise of not over-fitting, the model with higher complexity and finer spatial resolution can better describe the non-uniformity of the actual geological body, and thus is more physically close to the true situation and has higher credibility. The specific evaluation criteria are as follows:

[0125] Homogeneous velocity model: This model simplifies the entire monitoring area as a homogeneous medium, only containing a unified velocity value. This model has the lowest complexity and cannot reflect any vertical or horizontal velocity variation, and its applicability is highly dependent on the scale of the monitoring range. Therefore, its credibility is rated as low, and its quantitative credibility is directly set to 0.3. It can only be considered as a rough approximation in specific scenarios where the monitoring area is extremely small and the geological structure is extremely simple.

[0126] ​One-dimensional layered velocity model: This model has moderate complexity, assuming that velocity only varies with depth, which can effectively depict the layered sedimentary sequence of the strata. Its credibility is positively correlated with the fineness of stratification. For example, within the same depth range, a model that divides 1 km depth into 10 layers can better depict the vertical gradient of velocity than a model that only divides 2 layers, thus obtaining a higher credibility score. The credibility of such a model is rated as medium to high, which can be quantified as a value in the interval [0.4, 0.7] according to the rationality and integrity of the number of layers. If the number of layers is ≥5, it is 0.7, and if it is between 2 and 5, it is linearly quantified in the interval [0.4, 0.7].

[0127] Three-dimensional structural velocity model: This model has the highest complexity, can represent the anisotropy and non-uniformity of velocity distribution in three-dimensional space, and best reflects complex geological structures. Its credibility further depends on the fineness of grid discretization. The smaller the grid spacing, the stronger the model's ability to resolve complex structures (such as faults and collapse columns), and the higher its credibility. The credibility of such a model is rated as high, usually quantified as a value in the interval [0.8, 1.0]. Specifically, it can be refined as follows:

[0128] 0.80: The grid spacing is relatively large (> 50 m), which can only reflect regional-scale velocity changes and is difficult to analyze local complex structures; the model relies mostly on sparse observations.

[0129] 0.85: The grid spacing is moderate (30-50 m), which can basically depict large structures (such as major fault zones and thick lithological boundaries), but has limited sensitivity to small or subtle structures.

[0130] 0.90: The grid spacing is relatively small (15-30 m), which can better resolve medium-scale complex bodies (secondary faults and collapse columns), and the velocity field has a high degree of fitting with the measured data.

[0131] 0.95: The grid spacing is fine (5-15 m), which can clearly depict most complex structures, and the model is highly consistent with the inversion / observation data, with stable and reliable positioning and wavefield simulation results.

[0132] 1.00: The grid spacing is extremely fine (< 5 m), which is highly consistent with the actual geology on a full scale, can resolve minor structural changes, and achieves an ideal state; usually only under high-density observation and strong constraint conditions can it be achieved.

[0133] It is worth emphasizing that the credibility assessment of velocity model is not conducted in isolation, but in combination with the characteristics of input data and the requirements of positioning tasks. For travel-time positioning methods, when the input data is single-component waveform and the spatial scale of monitoring range is small, the velocity variation of the medium may not be significant. In this specific scenario, even if a velocity model with low credibility is used, the simplification error it brings may be within the engineering tolerance range, so this method can still be used as a viable option for rapid positioning. The decision framework will integrate the characteristics of "velocity model credibility", "data type identification", "effective station number", etc. to dynamically assess the applicability of different positioning methods in the current context. According to the Figure 3 The velocity model credibility index obtained by the velocity model of the first method is 0.6.

[0134] 3. Method library construction and applicability constraints

[0135] The system is built-in with three types of mainstream positioning methods: travel-time inversion (Powell), waveform migration stacking (QM_QuakeMigrate), and Bayesian inference (MaxEnt_HMC), and sets clear admission thresholds for them:

[0136] (1) Travel-time inversion positioning method (Powell positioning)

[0137] The Powell method is a traditional positioning algorithm based on seismic wave travel-time information. Its core idea is to minimize the residual error between observed travel-time and theoretical travel-time, and use optimization algorithms to invert the source location of mine earthquake events. This method relies on the arrival time information of the P-wave first arrival recorded by each seismic station, and through the construction of travel-time residual function and iterative adjustment of source parameters, the solution that minimizes the residual error is finally found.

[0138] The method uses a local optimization algorithm for parameter inversion, which has the advantage of fast solution speed, but due to its nature as a local search algorithm, it is sensitive to the selection of initial values, and when the parameter space is complex or the initial model deviates far from the true value, it is easy to fall into a local optimal solution. In terms of data requirements, this method is usually suitable for single-component (single-channel) data, and requires high signal-to-noise ratio and accurate first arrival picking to ensure the reliability of the travel time residual calculation. The access conditions of this method are relatively strict, which need to meet N_eff ≥ 4 (effective station number), SNR_mean ≥ 6 (average signal-to-noise ratio), σt_mean < 0.04 seconds (first arrival picking uncertainty), azimuth deviation G ≤ 180°, and velocity model confidence V_conf ≥ 0.5, to ensure that the input data reaches a high standard in terms of sample size, signal-to-noise level, travel time accuracy, and azimuth consistency. In terms of applicable scenarios, Powell is more suitable for observation environments with high data quality, reasonable station distribution, and clear first arrival, and it performs well in situations where high-precision single-point positioning is required and computing resources are limited. However, this method is sensitive to the rationality of station geometry and the accuracy of first arrival picking, and when the station distribution is poor or the first arrival identification has errors, the positioning effect may be greatly affected.

[0139] (2) QM method (full waveform migration stacking positioning)

[0140] The migration stacking method is based on the idea of diffraction stacking in exploration seismology. The main principle is to move and stack the waveforms received by the detectors along the one-way travel time, and to enhance the signal by stacking the waveforms at the potential source location, thereby determining the final source location. As shown in Figure 9 .

[0141] The core technology lies in calculating the migration stacking function , which accumulates the potential source location source and the signal offset of each detector. The formula is as follows:

[0142]

[0143] In the formula: represents the amplitude of the i-th wave; is the travel time between the source and the i-th detector; is the migration stacking amount, which is used to correct the migration time of each detector signal record; M is the number of waveforms.

[0144] Finally, the source location and the time of the earthquake are determined by maximizing all times t:

[0145]

[0146] The above is the flow of Kirchhoff-type migration stacking method, also known as diffraction stacking, which searches for both source location and shooting time simultaneously.

[0147] In terms of data requirements, the full waveform migration stacking method has relatively relaxed requirements for data quality, allowing for some first arrival picking errors, and supporting multi-component data input, further broadening its applicable observation scenarios. The access conditions of this method are relatively relaxed, only need to meet N_eff ≥ 3 (effective monitoring station number), SNR_mean ≥ 4 (average signal-to-noise ratio), and V_conf ≥ 0.50 (velocity model reliability), then the positioning calculation can be performed, which reflects its low running threshold and strong practicality.

[0148] The reason why the QM method can still achieve effective positioning with only three stations is that its waveform migration and stacking technology cleverly implies and utilizes time information. Instead of directly relying on the absolute arrival time difference in traditional geometric positioning, this method scans the continuous waveform data recorded by multiple stations in time and space: through the pre-calculated velocity model, the system applies the theoretical travel time of each possible source location to the waveform data of each station, and performs time migration; when the true source location is scanned, the waveform signals of each station after time migration will be maximally aligned and stacked (coherent), thus producing a significant peak in the coalescence function, and this peak corresponds to the position and time of the source. Therefore, even with only three stations, this method can extract the implicit high-precision time constraint by exploiting the coherence of the waveform itself during migration and stacking, thus achieving positioning.

[0149] In terms of applicable scenarios, migration stacking positioning is particularly suitable for seismic observation environments where the station layout is not ideal, the first arrival wave recognition has errors, or the overall data quality is at a medium level, and can significantly improve the adaptability to complex observation conditions while ensuring the reliability of the positioning results. It is a robust and practical earthquake positioning method.

[0150] (3) MaxEnt_HMC method (Bayesian method positioning)

[0151] The basic principle of MaxEnt-HMC is to combine Bayesian inference with Hamiltonian Monte Carlo (HMC) method, and introduce the maximum entropy principle in the sampling process to maximize the amount of information. In traditional HMC, by introducing momentum variables and simulating Hamiltonian dynamics, high-dimensional parameter space can be efficiently explored, but problems such as strong sample autocorrelation or easy to fall into local optimum may still occur. MaxEnt-HMC introduces negative log-likelihood and entropy terms in the objective function, so that the sampling process can maintain maximum uncertainty while fitting the observed travel time residuals, thereby avoiding excessive reliance on prior assumptions and improving robustness.

[0152] In terms of algorithm characteristics, this method uses Hamiltonian Monte Carlo (HMC) sampling technology, which can efficiently explore high-dimensional parameter space and effectively handle complex correlations between parameters, ensuring that the sampling process is more stable and efficient, and ultimately providing complete and statistically meaningful posterior distribution. In terms of data requirements, the Bayesian inference positioning method has the highest overall quality requirement for input data, which must meet N_eff ≥ 4 (effective sample size), SNR_mean ≥ 6 (average signal-to-noise ratio), and σt_mean < 0.04 seconds (average travel time residual), and V_conf ≥ 0.50 (velocity model confidence) to ensure that the sampling process can converge smoothly and obtain reliable results with statistical significance. The access conditions of this method are relatively strict, and due to the involvement of complex probability reasoning and sampling calculation of high-dimensional parameter space, its calculation cost is relatively high, so it is more suitable for running in an environment with sufficient computing resources. In terms of applicable scenarios, the Bayesian inference positioning method is particularly suitable for high-quality observation environments, especially in scientific research and high-precision application scenarios that require fine assessment of source location uncertainty, development of source mechanism inversion, or statistical analysis of complex earthquake sequences. This method can provide more comprehensive, reliable, and quantitative data support for subsequent seismological research, risk assessment, and emergency decision-making, but its high computational complexity also determines that it is more suitable for specialized scenarios with high requirements for positioning accuracy and statistical integrity.

[0153] When used, the above-mentioned methods can be positioned, in addition, the access conditions of Powell positioning algorithm must meet the HMC positioning algorithm, and more meet the access conditions of QM positioning algorithm. In the above-mentioned methods, the user wants to select the positioning result with high precision or use the least time to position. If you want to use the least time to position, you can select the positioning method according to the suggestion given in the minimum time delay mode; if the user wants to get high-precision positioning result, select the recommended positioning method in the precision priority mode. If the user wants to use short time and high precision, you can select the recommended positioning method in the comprehensive mode.

[0154] In short, according to the actual input waveform data, station coordinate information, velocity model and other information, the access conditions are determined which positioning method is suitable, and then the user's own demand is selected.

[0155] 4. Rule engine and strategy decision

[0156] Based on the above index, the system makes dynamic decision through the rule engine, and outputs single method or cascade strategy. The core rules include:

[0157] (1) Minimum time delay mode: prefer Powell, then fall back to QM or MaxEnt_HMC, if it is three-component data, then disable Powell.

[0158] (2) Precision priority mode: if cascade is allowed and Powell and MaxEnt_HMC meet the requirements, prefer to use Powell to MaxEnt_HMC, otherwise try to use QM to MaxEnt_HMC, and finally use single MaxEnt_HMC positioning method as fallback.

[0159] (3) Need uncertainty mode: prefer to guarantee MaxEnt_HMC participation (cascade mode or single MaxEnt_HMC), which is used to output posterior and uncertainty quantification. Cascade means running a fast positioning method (Powell or QM) first to get the initial solution or coherent zone, and then running MaxEnt_HMC to get the posterior distribution, uncertainty and refinement result, which takes into account speed and robustness

[0160] (4) Comprehensive mode: limitedly select Powell when Powell meets the requirements and no posterior is needed, otherwise prefer to use QM; select QM to MaxEnt_HMC when posterior is needed and cascade is allowed.

[0161] 5. Explainable output and system flow

[0162] After the system decision is completed, the interpretable output and report generation link is entered. This module aims to present the decision logic, key indicator evaluation results, and the final recommended positioning strategy inside the system to the user in a structured and highly readable manner, significantly enhancing the transparency, operability, and user trust of the system. Specifically, this module generates a detailed decision report, which typically includes the following parts:

[0163] (1) Event basic information: Clearly list the unique number of the processed mine earthquake event, occurrence time, total number of participating stations, and other basic metadata to provide clear event information for the user.

[0164] (2) Key indicator summary and evaluation: Display the core evaluation parameters calculated by the system in a quantitative form, such as the number of effective stations (Neff), average signal-to-noise ratio (SNR_mean), average pick uncertainty (σt_mean), average maximum station gap angle (AG), velocity model confidence (V_conf), data type (single component / three-component), and current computing resource status, etc.

[0165] (3) Step-by-step explanation of the decision process:

[0166] A) First, explain whether the input data is in three-component data format, and determine whether to use the travel time positioning method;

[0167] B) List the admission conditions of each candidate positioning method and its satisfaction one by one ("Powell method: effective station number ≥ 4, average SNR ≥ 6.0, σt_mean < 0.04 seconds (initial pick uncertainty), azimuth angle deviation G ≤ 180°, velocity model confidence V_conf ≥ 0.5"; "MaxEnt_HMC method: effective station number ≥ 4, σt_mean < 0.04 s, velocity model confidence V_conf ≥ 0.5, average SNR ≥ 6 dB"; "QM method: effective station number ≥ 3, average signal-to-noise ratio ≥ 4 dB, and velocity model confidence V_conf ≥ 0.50"

[0168] C) Explain how to make the best trade-off and selection of the methods that pass the screening according to the user's preset or system's default strategy mode ("In the 'precision first' mode, although Powell and MaxEnt_HMC both meet the admission conditions, the cascade strategy QM→MaxEnt_HMC is preferentially recommended to balance computing efficiency and posterior uncertainty quantification").

[0169] (4) Final recommended scheme: Clearly give the recommended positioning method or cascade strategy of the system. If it is a cascade strategy, it will explain its execution process, and the report will also indicate the expected output type of the scheme.

[0170] (5) Conclusion and recommendations: The final part summarizes the decision-making process and provides supporting evidence.

[0171] 6. Blast event location verification

[0172] Considering the trade-off between accuracy and computational time, the QM method is recommended. If accuracy is prioritized, the QM + MaxEnt_HMC cascade strategy is recommended. We will use the Powell method, QM method, MaxEnt_HMC method, and QM + MaxEnt_HMC cascade strategy to locate this blast event in turn. The positioning error and computational time are as shown in Table 2:

[0173] Table 2: Summary of blast event location results using different strategies

[0174]

[0175] From the results in Table 2, the feasibility of the positioning decision-making method proposed in this invention can be seen. In terms of positioning accuracy, the QM→MaxEnt_HMC cascade strategy has the highest accuracy, followed by QM, then Powell, and the worst is MaxEnt_HMC. The main reason for the difference in accuracy is that the results of MaxEnt_HMC are highly dependent on the setting of the initial value, and the selection of the initial value has a greater impact on its accuracy.

[0176] In terms of time consumption, the Powell method is the fastest, followed by MaxEnt_HMC, then QM, and the QM→MaxEnt_HMC cascade is the slowest.

[0177] Therefore, when choosing a balanced strategy, selecting QM is reasonable because it ranks second in accuracy and its time consumption, although ranking third, is actually closely related to the setting of the grid size. By adjusting the grid resolution, the efficiency of QM can be optimized

[0178] For accuracy-first strategies (i.e., using QM→MaxEnt_HMC), this combination is also effective. QM provides a relatively accurate initial value, and MaxEnt_HMC only needs to sample in a smaller area based on this value, thereby obtaining a more accurate positioning result. It is worth noting that the total time consumption of QM + MaxEnt_HMC combination is not equal to the sum of the time consumption when they are executed independently. In fact, since QM has effectively reduced the sampling range of MaxEnt_HMC, the final total time consumption is lower than the sum of the two independent executions.

Claims

1. A multi-index fusion discriminant mine earthquake adaptive positioning system, characterized in that: The system comprises a data acquisition module, a feature extraction and quality evaluation module, a comprehensive index calculation module and an adaptive decision module. The data acquisition module is used for receiving and pre-processing seismic event waveform data of multiple stations, station spatial coordinate information and velocity model parameter information, and outputting waveform data and metadata in a unified format to the feature extraction and quality evaluation module. Then, the P-wave arrival time is obtained through the automatic phase picking algorithm of the module, and the signal-to-noise ratio and picking uncertainty of each station are calculated, and the results are output to the comprehensive index calculation module. After receiving the primary parameters from the feature extraction and quality evaluation module, the comprehensive index calculation module performs comprehensive calculation to generate a multi-dimensional feature vector including the number of effective stations, the average signal-to-noise ratio, the average picking uncertainty, the station gap angle, the velocity model reliability, the data type identification and the resource availability, and outputs the feature vector to the adaptive decision module. Through the rule engine and positioning method library built in the adaptive module, combined with the multi-dimensional feature vector received from the comprehensive index calculation module, the optimal positioning strategy is dynamically selected or combined according to the preset rules and decision modes, and the final positioning scheme is output.

2. A mine earthquake adaptive positioning method using the multi-index fusion discrimination of the mine earthquake adaptive positioning system of claim 1, characterized in that, The steps are as follows: Step 1, system initialization and data preprocessing: the system receives multiple station seismic waveform data and pre-processes the data. Based on the pre-processed waveform data, the system uses classical signal detection and picking algorithm to automatically pick P-wave first arrival of each station waveform, and synchronously calculates the signal-to-noise ratio SNR and picking uncertainty σt of each station; Step 2, key parameter extraction and quality evaluation: based on the data results in step 1, the key indicators for evaluating the overall quality and applicability of the current data set are calculated, including: the number of effective stations, the signal-to-noise ratio, the picking uncertainty, the maximum station gap angle, and the velocity model reliability; Step 3, rule engine and strategy decision: the system is built in with the travel time inversion positioning method Powell, the waveform offset stacking positioning method QM and the Bayesian positioning method MaxEnt_HMC. Based on the indicators in step 2, the system makes dynamic decisions through the rule engine, and outputs single method or cascade strategy; Minimum time delay mode: first, if the input data is three-component data, the system directly disables the Powell method; if the Powell method is not disabled and its admission condition is met, the single Powell positioning method is preferentially selected; otherwise, if the admission condition of the QM method is met, the single QM positioning method is selected; otherwise, if the admission condition of the MaxEnt_HMC method is met, the single MaxEnt_HMC positioning method is selected. Precision-first mode: if both Powell and MaxEnt_HMC methods are allowed to be cascaded and their admission conditions are satisfied, the system recommends and executes Powell → MaxEnt_HMC cascaded localization strategy; otherwise, if both QM and MaxEnt_HMC methods are allowed to be cascaded and their admission conditions are satisfied, the system recommends and executes QM → MaxEnt_HMC cascaded localization strategy; otherwise, if the admission condition of MaxEnt_HMC method is satisfied, the system recommends and executes single MaxEnt_HMC localization method as fallback; Uncertainty-first mode: the system prioritizes the goal of ensuring the output of complete posterior distribution and uncertainty quantification, and the decision logic is as follows: if both Powell and MaxEnt_HMC methods are allowed to be cascaded and their admission conditions are satisfied, the system recommends and executes Powell → MaxEnt_HMC cascaded strategy, which first uses the Powell method to quickly obtain a high-precision initial source parameter, and then uses the MaxEnt_HMC method to sample in the neighborhood of this initial solution to efficiently obtain a refined posterior distribution and uncertainty; otherwise, if both QM and MaxEnt_HMC methods are allowed to be cascaded and their admission conditions are satisfied, the system recommends and executes QM → MaxEnt_HMC cascaded strategy, which first uses the QM method to determine a high-probability source spatial coherence region, and then uses the MaxEnt_HMC method to sample in this coherence region to robustly obtain the posterior distribution and uncertainty; otherwise, if the admission condition of MaxEnt_HMC method is satisfied, the system directly recommends and executes single MaxEnt_HMC localization method to output the posterior distribution and uncertainty; Comprehensive mode: in this mode, the system makes dynamic decisions according to the user's requirements for the posterior distribution, and the logic is as follows: a. When the user does not require the output of posterior distribution and uncertainty, if the admission condition of Powell method is satisfied, the system prioritizes and executes single Powell localization method to achieve fast localization; otherwise, if the admission condition of QM method is satisfied, the system recommends and executes single QM localization method; otherwise, if the admission condition of MaxEnt_HMC method is satisfied, the system recommends and executes single MaxEnt_HMC localization method; b. When the user requires the output of posterior distribution and uncertainty, if the admission conditions of both QM and MaxEnt_HMC methods are satisfied, the system prioritizes and executes QM → MaxEnt_HMC cascaded localization strategy, which first uses the QM method to determine a high-probability source coherence region, and then uses the MaxEnt_HMC method to sample in this region to balance efficiency and reliability of uncertainty quantification; otherwise, if the admission condition of MaxEnt_HMC method is satisfied, the system recommends and executes single MaxEnt_HMC localization method. Step 4, after the system decision is completed, enter the interpretation output and report generation link, output the decision logic inside the system, the key index evaluation result and the final recommended positioning strategy, and generate a decision report.

3. The multi-index fusion discriminant mine earthquake adaptive positioning method according to claim 2, characterized in that: The specific method in step 1 is: Step 1.1, the system receives multi-station seismic waveform data in single-component and three-component formats, station information files containing spatial coordinates and component type metadata, velocity model parameters including model type, grid spacing and calibration residual, and grid parameters including spatial resolution; Step 1.2, determine whether the waveform recorded by each station is single-component or three-component; for three-component data, the system preferentially uses the Z component for subsequent processing, and if there is no Z component, it generates an equivalent single-component waveform through EN energy stacking; Step 1.3, based on the preprocessed waveform data, use classical signal detection and picking algorithms to automatically pick the P wave first arrival of each station waveform; The STA / LTA method is based on the significant change in energy characteristics of seismic signals before and after the arrival of P waves. The first arrival detection is realized by calculating the ratio of the average energy of the sliding time window. Its formula is expressed as: ； ； where is the waveform amplitude, and are the short and long time window length respectively, the AIC method segments the waveform by a local autoregressive model, and finds the point that minimizes the information as the phase arrival time, which is expressed as ； Calculate the short-time window / long-time window of the waveform signal, that is, the STA / LTA amplitude average ratio. When the ratio exceeds the preset threshold, it is determined that the P wave first arrival has arrived, and the preliminary arrival time point is obtained; In the time window near the first arrival triggered by STA / LTA, the signal is further analyzed using the Akaike Information Criterion (AIC) to determine the optimal position of the waveform feature point, achieving P wave arrival time picking. The ratio method trigger combined with AIC refinement algorithm for automatic P wave arrival time picking, simultaneously calculates the signal-to-noise ratio (SNR) and picking uncertainty (σt) of each station.

4. The multi-index fusion discriminant mine earthquake adaptive positioning method according to claim 2, characterized in that: The specific method in step 2 is: Step 2.1, signal-to-noise ratio calculation and effective station number The signal-to-noise ratio is defined as the logarithmic form of the ratio of the signal energy in the P wave window to the energy in the noise window: ； Where: the picking uncertainty is estimated based on the curvature of the AIC function near the arrival time point or the standard deviation of multiple picking results, Define the stations with SNR≥5.0 and σt≤0.04s as effective stations, only select the effective stations for positioning, the average signal-to-noise ratio is the average of the SNR values of all effective stations, the average picking uncertainty is the average of the P wave arrival time picking uncertainty σt of all effective stations, reflecting the reliability of the overall P wave arrival time picking, the smaller σt, the more reliable the timing basis for positioning; Step 2.2, analysis of station gap angle and construction of heat map The station gap angle AG represents the maximum angle blank area formed by the station direction vector in the horizontal plane projection of any point in the monitoring area, that is, the maximum uncovered angle between adjacent station azimuths, reflecting the geometric coverage of the station to the source point in the plane: the smaller the gap angle, the more uniform the station coverage, the higher the positioning accuracy; on the contrary, if the gap angle is large, there is a significant monitoring blind area, which can easily lead to increased source positioning error or missed detection; The specific calculation process is as follows: For any source point within the monitored region , the direction vector to each station is first computed: ； Subsequently, each direction vector is converted to an azimuth angle : ； Sort the azimuths of all stations in ascending order, calculate the included angle between the adjacent two azimuths: ； The angle between the last station and the first station is: ； Finally, the maximum value of all the angles is taken as the maximum gap angle at the hypocenter: ； Through the calculation and statistics of the AG values of potential hypocenter positions in the monitoring area of the three-component station envelope, a two-dimensional AG heat map is generated to show the uniformity of station direction coverage at different positions in the monitoring area; Define that when the average gap angle ≤90°, the spatial distribution of the station is better, and the monitoring coverage is more sufficient; if the regional gap angle >270°, it indicates that there is a significant lack of monitoring direction at this position, and by adding stations or adjusting the layout of existing stations, the monitoring blind area can be eliminated; Step 2.3, velocity model credibility analysis The credibility of the velocity model is used to quantify the priori index of the characterization ability of the current velocity model to the true underground wave velocity structure. This index is based on the complexity and spatial resolution of the model for comprehensive evaluation. The core principle is: under the premise that the model is effectively constrained, the higher the complexity and the finer the spatial resolution of the model, the stronger its ability to depict the actual geological heterogeneity, and thus it is given a higher credibility; The following quantitative evaluation criteria are used by the system to assign credibility values to the input velocity model: Step 2.3.

1. Mean velocity model The entire monitoring area is simplified as a homogeneous medium, containing only one uniform velocity value; this model has the lowest complexity, and its credibility is fixed at 0.3; Step 2.3.

2. One-dimensional layered velocity model Assuming that the velocity model only changes with depth, it can depict the layered sedimentary sequence of the strata. The credibility of this model is linearly quantified according to the number of layers: when the number of layers ≥5, V_conf = 0.7; when 2≤number of layers <5, V_conf = 0.4+0.1×(number of layers-2); when the number of layers <2, it is classified as a mean model, and the credibility is 0.3; Step 2.3.

3. Three-dimensional structure velocity model This model can represent the anisotropy and non-uniformity of velocity distribution in three-dimensional space; the credibility of the model is determined by the grid spacing: if the grid spacing >50m, V_conf = 0.7; if 30m < grid spacing ≤50m, V_conf = 0.8; if 15m < grid spacing ≤30m, V_conf = 0.9; if 5m < grid spacing ≤15m, V_conf = 0.95; if grid spacing ≤5m, V_conf = 1; The velocity model credibility V_conf is an independent quantitative index, which is input into the rule engine of the adaptive decision module together with "data type identification, single / three-component", "effective station number N_eff", and other dimensional characteristics; The rule engine will comprehensively consider all these parameters and evaluate the applicability of different positioning methods according to the pre-set and determined rules.

Images