Intelligent source localization method for microseismic earthquakes, model training method and system

Abstract

This application proposes an intelligent microseismic source location method, a model training method, and a system. This method incorporates physical laws such as the seismic wave equation, comprehensively considers physical loss and data loss, improves the model's adaptability to complex geological environments, effectively addresses the challenges of microseismic source location in complex geological environments, improves noise resistance and location accuracy, and provides more reliable technical support for the safety monitoring of underground engineering.

Description

Technical Field

[0001] This application relates to the field of mine safety monitoring technology, and in particular to a method for intelligent localization of microseismic sources, a method for training models, and a system. Background Technology

[0002] Microseismic monitoring has broad application prospects in fields such as mining, earthquake prediction, and shale gas extraction. Its principle is to receive microseismic data from sensors and accurately locate the location and time of the seismic source event based on the microseismic data.

[0003] Currently, traditional microseismic source location methods mainly rely on seismic wave propagation theory, such as the Geiger method, grid search method, and double-difference method. These methods are generally based on accurate velocity models and are quite sensitive to the selection of initial points. In complex geological environments, the propagation velocity of seismic waves may change due to the inhomogeneity of the subsurface medium, leading to a decrease in the location accuracy of traditional methods. Furthermore, in noisy environments, the location accuracy of the algorithms is affected, causing location deviations. Therefore, traditional microseismic source location methods still have certain limitations. Summary of the Invention

[0004] This application proposes an intelligent microseismic source localization method, a model training method, and a system, which can solve one of the problems existing in the background technology.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] Firstly, a training method for a microseismic source location model is provided, the training method comprising:

[0007] Training data is obtained, wherein the training data is time-related microseismic waveform data obtained by sensor sensing; and

[0008] The training data is used to train a microseismic source location model, wherein the microseismic source location model adopts a physical information neural network (PINN), and the PINN takes the microseismic waveform data as input and the spatial coordinates of the source and the time of occurrence as output.

[0009] Based on the above technical solution, this method incorporates physical laws such as the seismic wave equation, comprehensively considers physical loss and data loss, improves the model's adaptability to complex geological environments, effectively addresses the challenges of microseismic source location in complex geological environments, improves noise resistance and positioning accuracy, and provides more reliable technical support for the safety monitoring of underground engineering.

[0010] In one possible design approach of the first aspect, the overall loss function adopted by the PINN includes a data loss term and a physical loss term, wherein the physical loss term is defined by a wave equation.

[0011] In one possible design approach of the first aspect, the wave velocity parameter in the wave equation is determined by one of the following methods: obtaining the equivalent wave velocity through first arrival wave travel time inversion, the average wave velocity of a preset geological region, and the spatially correlated wave velocity field obtained by neural network inversion.

[0012] In one possible design approach of the first aspect, the data loss term is defined by predicting the source displacement and the actual measured displacement.

[0013] In one possible design approach of the first aspect, the overall loss function employs an adaptive weighting strategy.

[0014] In one possible design approach of the first aspect, the overall loss function L total for:

[0015] Where Ldata is the data loss term, Lphys is the physical loss term, α is the weight of the data loss term, λ is the weight of the physical loss term, i is the data point index, N is the total number of sampling points, j is the physical constraint index, and M is the total number of physical constraint points. For the predicted medium response displacement, This represents the actual measured displacement of the medium. The spatial location of the i-th data point Let i be the sampling time for the i-th data point. Let j be the spatial location of the j-th data point. Let c be the sampling time of the j-th data point, and c be the wave velocity. is the Laplace operator, representing the change in wave propagation in space, and k is the number of training iterations.

[0016] In one possible design of the first aspect, the training method further includes: batch normalizing the microseismic waveform data, wherein the batch normalization specifically includes:

[0017] Based on the time-frequency analysis characteristics of the microseismic waveform data, the sliding window type is determined; and

[0018] When the time-frequency analysis characteristics of the microseismic waveform data indicate that the microseismic waveform data is a high-frequency transient signal, a short window is used for batch normalization; when the time-frequency analysis characteristics of the microseismic waveform data indicate that the microseismic waveform data is a low-frequency continuous signal, a long window is used for batch normalization.

[0019] Secondly, a method for intelligent localization of microseismic sources is provided, the method comprising:

[0020] Obtain the current microseismic waveform data; and

[0021] The current microseismic waveform data is input into the microseismic source localization model trained as described above, and the current source spatial coordinates and current time of occurrence are obtained as output.

[0022] Thirdly, a microseismic source intelligent positioning system is provided, the system comprising:

[0023] The sensor module is used to obtain the raw data of the current micro-vibration waveform;

[0024] The preprocessing module is used to standardize the raw data of the current microseismic waveform to obtain the current microseismic waveform data.

[0025] The intelligent computation and positioning module is used to call the PINN trained by the above training method to process the current microseismic waveform data to obtain the current spatial coordinates of the seismic source and the time of occurrence; and

[0026] The visualization module is used to generate and store dynamic wavefront propagation diagrams based on the current spatial coordinates of the seismic source.

[0027] In one possible design of the third aspect, the sensor module includes:

[0028] A three-component high-precision probe using an orthogonally arranged sensor array;

[0029] The calibrated rotating component is linked with the three-component high-precision probe to support remote control of triaxial orientation calibration.

[0030] An edge storage chip is directly connected to the three-component high-precision probe for temporary storage of monitoring data; and

[0031] The BeiDou timer is connected to the edge storage chip via an optical fiber composite cable to ensure that the data timestamp is synchronized with the ground system.

[0032] Fourthly, an electronic device is provided, comprising: a processor, and a memory coupled to the processor, the memory for storing a computer program; the processor for executing the computer program stored in the memory such that the electronic device performs the training method as described in any possible implementation of the first aspect, or performs the positioning method as described in the second aspect.

[0033] Fifthly, a computer-readable storage medium is provided, including a computer program or instructions that, when executed on a computer, cause the computer to perform the training method as described in any possible implementation of the first aspect, or to perform the positioning method as described in the second aspect.

[0034] A sixth aspect provides a computer program product comprising: a computer program or instructions that, when executed on a computer, cause the computer to perform the training method as described in any possible implementation of the first aspect, or to perform the positioning method as described in the second aspect. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a schematic diagram of a microseismic source localization method and system based on Physical Information Neural Network (PINN) provided in an embodiment of this application.

[0037] Figure reference numerals: 1-Data acquisition module, 101-Microseismic sensor array, 102-Data acquisition device, 103-Data storage unit, 2-Data preprocessing module, 201-Noise reduction, 202-Signal normalization, 203-Feature extraction, 3-Intelligent computing and positioning module, 301-Microseismic wave propagation equation modeling, 302-Source physical law constraint, 303-Neural network architecture, 304-Loss function optimization, 305-Adaptive optimization framework, 306-Microseismic source positioning inference, 4-Result analysis and visualization module, 401-Error analysis, 402-3D visualization, 403-Data output interface.

[0038] Figure 2 This application provides a microseismic waveform diagram of a microseismic source localization method and system based on Physical Information Neural Network (PINN).

[0039] Figure 3This is a detailed diagram of the PINN network structure of a microseismic source localization method and system based on Physical Information Neural Network (PINN) provided in an embodiment of this application.

[0040] Figure labels: 5-Input layer, 6-Hidden layer, 7-Output layer, 8-Automatic differentiation, 9-Loss function (data loss and physical regularization term loss), 10-Minimize loss function, 11-Gradient descent, 12-Model weights, 13-Weight update.

[0041] Figure 4 This is a schematic diagram of a microseismic source localization method and system simulation based on a Physical Information Neural Network (PINN) provided in an embodiment of this application.

[0042] Figure labels: 14-sensor, 15-seismic source.

[0043] Figure 5 This application provides a flowchart of a microseismic source localization method and system based on Physical Information Neural Network (PINN).

[0044] Figure labels: 16-Sensor array, 17-Data flow direction, 18-Underground station, 19-Ground control center, 20-Data preprocessing module, 21-Intelligent computing and positioning module, 22-Result analysis and visualization module, 23-Mine working face.

[0045] Figure 6 This is a schematic diagram of the underground detection sensor of a microseismic source localization method and system based on Physical Information Neural Network (PINN) provided in an embodiment of this application.

[0046] Reference numerals: 24-Three-component high-precision probe, 25-Composite explosion-proof housing, 26-Calibration rotation assembly, 27-Indicator light, 28-Edge storage chip, 29-Backup power supply, 30-BeiDou timer, 31-(External) fiber optic composite cable.

[0047] Figure 7 This is a schematic diagram of a microseismic source localization method and system control center based on Physical Information Neural Network (PINN) provided in an embodiment of this application.

[0048] Reference numerals: 32-Signal receiving device (shallow wireless signal), 33-Preprocessing chassis, 34-Signal indicator light, 35-Fiber optic composite cable, 36-Intelligent computing module box, 37-Neural network computing chip matrix, 38-Heat dissipation hole, 39-Display screen, 40-Display waveform diagram. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0050] It should be noted that although functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification and the above-mentioned figures are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0052] The following provides an exemplary description of the microseismic source location method, training method, and system according to embodiments of this application. It is understood that the location method and training method are largely the same in terms of processing steps.

[0053] This application proposes a method and system for intelligent microseismic source localization based on a Physics-Informed Neural Network (PINN). This method incorporates physical laws such as the seismic wave equation, comprehensively considering both physical and data losses to improve the model's adaptability to complex geological environments. Compared with traditional methods, this approach effectively addresses the challenges of microseismic source localization in complex geological environments, improves noise resistance and localization accuracy, and provides more reliable technical support for the safety monitoring of underground engineering projects.

[0054] like Figure 1 As shown, the PINN-based microseismic source localization system provided in this embodiment includes:

[0055] The data acquisition module (1) is responsible for acquiring the raw waveform data of microseismic events. This module uses a high-sensitivity sensor array to acquire seismic wave data in real time and convert it into a computer-processable format. The goal of data acquisition is to ensure that accurate raw waveform information is acquired for subsequent analysis and processing.

[0056] The data acquisition module (1) includes: a micro-vibration sensor array (101) for acquiring micro-vibration signals, with multiple high-frequency sensors deployed in the monitoring area and a sampling frequency of 6000Hz to ensure high spatiotemporal resolution; a data acquisition device (102) for performing analog-to-digital conversion (A / D conversion) on the analog signals acquired by the sensors and for performing signal conditioning to improve the signal-to-noise ratio; and a data storage unit (103) for storing the original waveform data and supporting data transmission to the subsequent calculation unit.

[0057] The data acquisition device (102) uses a high-precision analog-to-digital converter chip, which supports multi-level signal amplification and filtering to improve the signal-to-noise ratio of the micro-vibration signal.

[0058] The data storage unit (103) adopts a high-capacity storage medium and supports data compression and remote access functions. It also provides a distributed storage solution to improve data access efficiency and security, and facilitate long-term data storage and retrieval.

[0059] The data preprocessing module (2) preprocesses the acquired raw waveform data, such as denoising, normalization, and signal filtering, to improve data quality; removes background noise and other irrelevant signals to ensure the accuracy and consistency of the data and provide stable input to the subsequent positioning algorithm.

[0060] The data preprocessing module (2) includes: noise reduction (201), which filters the original signal to eliminate environmental noise and system errors; signal normalization (202), which standardizes the data from different sensors to eliminate differences in device response; and feature extraction (203), which uses time-frequency transformation (such as short-time Fourier transform) to extract key signal features to enhance source information.

[0061] Signal normalization employs the batch normalization method, specifically:

[0062] Based on the non-stationary characteristics of microseismic signals, the sliding window length is adaptively adjusted through time-frequency analysis. The window range is 300-12000 sampling points. A short window (300-1000 sampling points) is used for high-frequency transient signals to preserve detailed features, while a long window (5000-12000 sampling points) is used for low-frequency continuous signals to suppress local noise.

[0063] The determination of the dynamic time window is based on a multi-index driven strategy. The system first performs feature extraction analysis on the microseismic signal, extracting the signal duration, instantaneous energy change rate and waveform envelope curve as the basis for decision-making.

[0064] The duration is estimated by constructing an energy envelope using the envelope function E(t)=|H(x(t))| (where H represents the Hilbert transform), and determining the start and end boundaries of the signal based on its upper and lower thresholds (such as setting 10% of the maximum value), thereby estimating the duration of the vibration.

[0065] The instantaneous energy change rate is expressed by the formula R(t)=d(x²(t)) / dt, which reflects the local change gradient of signal power on the time axis. The waveform envelope curve is based on the above Hilbert transform result E(t), which identifies the main shock segment and aftershock segment, and extracts the peak concentration region through moving average and kurtosis analysis.

[0066] The intelligent computing positioning module (3) is based on the optimization strategy of physical loss function and data loss function, combined with the constraint of wave equation, to train neural network to infer the spatial coordinates of the earthquake source and the time of earthquake occurrence; optimize physical loss and data loss, significantly improve the accuracy and robustness of microseismic source positioning, and have stronger anti-noise ability; it can perform accurate microseismic source positioning in complex geological environments.

[0067] The intelligent computing positioning module (3) includes: micro-seismic wave propagation equation modeling (301), which constructs micro-seismic wave propagation equations based on elastic wave theory to constrain micro-seismic source positioning calculations; source physical law constraint (302), which replaces the traditional layered velocity model with equivalent wave velocity parameters and obtains the spatial average wave velocity through wave equation constraint inversion; neural network architecture (303), which includes an input layer (sensor coordinates, arrival time data), a hidden layer (deep neural network), and an output layer (seismic source position x, y, z, and seismic origin time t); loss function optimization (304), which adopts a hybrid constraint loss, including data-driven loss and physical regularization loss, to improve positioning accuracy and stability of micro-seismic source positioning; adaptive optimization framework (305), which combines the Adam optimization algorithm with gradient descent mechanism to adjust the learning rate and dynamically update the loss function weights to achieve adaptive training, improve convergence stability, avoid premature convergence, and improve the generalization ability of the network; and micro-seismic source positioning inference (306), which utilizes the trained PINN. The model infers the source coordinates and time of the earthquake, supports a GPU-accelerated parallel computing architecture to improve computational efficiency and meet the needs of large-scale data processing, and achieves the following processing: block-based parallel preprocessing of sensor data; synchronous backpropagation calculation of physical loss function and data loss function; and concurrent inference of multi-source events.

[0068] The results analysis and visualization module (4) is used to output the spatial coordinates of the seismic source, the time of the seismic source occurrence, and the results of the location error analysis. This module provides intuitive visualization tools to help users evaluate the location results of the microseismic source and supports the graphical interface to display the location and error information of the seismic source, which can assist engineering decision-making and subsequent analysis.

[0069] The results analysis and visualization module (4) includes: error analysis (401), which calculates the microseismic source location error (such as Euclidean error, residual analysis) to evaluate the model accuracy; three-dimensional visualization (402), which displays the source distribution based on a three-dimensional coordinate system to improve interpretability, and combines GIS (Geographic Information System) and mining area geological model to support the fusion of geological modeling data and achieve accurate spatial distribution mapping of the source; and data output interface (403), which supports exporting the results to Excel, database storage or other engineering application systems for subsequent analysis and decision-making.

[0070] The data acquisition module (1) belongs to the underground exploration part, while the data preprocessing module (2), intelligent computing and positioning module (3), and result analysis and visualization module (4) all belong to the ground control center part.

[0071] In the process of microseismic source location, the propagation characteristics of seismic waves must be accurately simulated. Seismic waves are generally classified into P-waves (longitudinal waves) and S-waves (transverse waves), and their propagation speed and mode vary under different geological conditions. To improve the accuracy of microseismic source location, the wave equation, as a fundamental physical model, is embedded into the training process of a neural network. The wave equation describes the propagation law of seismic waves in a medium and is usually expressed in the following form:

[0072] One-dimensional wave equation:

[0073] Where c is the wave velocity, x is the position, t is the time, and u(x,t) represents the displacement of the particle at position x in the medium at time t, that is, the vibration displacement of the medium particles caused by the propagation of elastic waves after the source excitation.

[0074] The wave velocity parameter c represents the propagation speed of seismic waves in a medium and is a core parameter for location calculations. This wave velocity parameter c can be determined in any of the following ways:

[0075] Equivalent wave velocity: The arrival time information of the first arrival wave recorded by the sensor is used to calculate the comprehensive wave velocity on the path from the source to each receiver by the least squares fitting travel time inversion method. It is used to represent the average characteristics of the actual propagation path and is suitable for situations where the wave velocity structure is unknown or computational resources are limited.

[0076] Average wave velocity: Based on existing geological exploration, a fixed wave velocity value is preset for the entire area. This method is suitable for scenarios with homogeneous media or small wave velocity variations, significantly simplifying the calculation process and is a commonly used simplified model in traditional positioning methods.

[0077] Wave velocity field: The spatially correlated wave velocity model c(x,y,z) obtained by neural network inversion can dynamically reflect the medium characteristics at different locations and is suitable for high-precision seismic source location under complex and non-uniform geological conditions.

[0078] The three wave velocity forms mentioned above are essentially specific implementations of parameter c in the wave equation. They can be flexibly selected according to the application scenario, taking into account both computational efficiency and positioning accuracy.

[0079] If it is multidimensional, then the wave equation is:

[0080] in, It is the Laplace operator, representing the change in the propagation of wave in space.

[0081] like Figure 3 As shown, PINN combines physical laws with data-driven optimization methods to optimize the loss function. The loss function consists of a data loss function and a physical loss function. The data loss function measures the error between the neural network's prediction and the actual measured data, while the physical loss function ensures that the neural network's output conforms to the laws of seismic wave propagation. PINN's loss function consists of data loss and physical loss (regularization term):

[0082] Data loss function:

[0083] Where i is the index of the data point (the i-th sampling point), and N is the total number of sampling points. Let i be the spatial location of the i-th data point. Let i be the sampling time for the i-th data point. The displacement of the medium response predicted by the neural network. This refers to the actual displacement data of the medium measured by the sensor.

[0084] Physical loss function:

[0085] Where j is the physical constraint index (the j-th sampling point), and M is the total number of physical constraint points. Let j be the spatial location of the j-th data point. Let be the sampling time for the j-th data point.

[0086] The loss function measures the residuals of the wave equation, ensuring that the source location output by the neural network conforms to the physical laws of seismic wave propagation.

[0087] The total loss function of PINN is obtained by weighted summation of the data loss function and the physical loss function, and the ultimate goal is to minimize the total loss function.

[0088] in, , It is a weighting coefficient used to balance the contributions of data loss and physical loss.

[0089] In traditional PINN, α and λ are fixed hyperparameters. However, in actual training, data loss dominates in the early stages, mainly to quickly fit the observed data, while physical loss dominates in the later stages, mainly to ensure physical consistency. To balance the loss contribution at different training stages and improve convergence speed and localization accuracy, an adaptive weighting strategy is proposed.

[0090] Where k is the number of training iterations, and the weights are dynamically adjusted based on the loss value of the previous round.

[0091] The location of a seismic source can be achieved by utilizing the time difference of arrival (TDOA) or the time of arrival (TOA). The basic formula for microseismic source location is:

[0092] Where n is the sensor index (the nth sensor). This is the excitation time of the seismic source, which is also the required time of earthquake occurrence. Let be the time it takes for the seismic source signal to reach the nth sensor, assuming the seismic source is located at... The sensor is located at .like Figure 4 As shown.

[0093] The solution approach is to minimize the time difference error of multiple sensors and then solve for the spatial coordinates of the earthquake source.

[0094] In terms of error calculation, to quantify the accuracy of microseismic source location, this invention introduces the following error analysis formula:

[0095] Absolute error:

[0096] Relative error:

[0097] in,( y s z s (x) represents the actual source coordinates. pred y pred z pred ( ) represents the predicted earthquake source coordinates.

[0098] Furthermore, the PINN microseismic source localization module combines physical information with neural network deep learning algorithms, enabling it to adaptively perform source inversion under various complex geological environments and accurately obtain the spatial coordinates and temporal information of the source.

[0099] The wave equation constraint module, based on seismic wave propagation theory and the physical properties of the propagation medium, uses the wave equation to constrain the solution space during model training, effectively improving the noise resistance of microseismic source location and adapting to various noise interferences and complex geological conditions.

[0100] Furthermore, the system also includes:

[0101] Sensor network module: Composed of multiple sensor nodes, this module collects microseismic signals and transmits them to the data acquisition module, ensuring real-time signal transmission and accuracy. Sensor nodes can utilize the BeiDou system for time synchronization to ensure data time synchronization.

[0102] Communication system module: Includes wired and wireless communication methods to ensure real-time data transmission and effective feedback, and supports large-scale monitoring and multi-node collaborative operation.

[0103] Furthermore, the data acquisition module incorporates denoising and normalization algorithms to ensure the high quality and consistency of the acquired data, significantly improving the stability and accuracy of subsequent algorithm processing.

[0104] Furthermore, the results analysis and visualization module has efficient data processing and graphical display functions, and can generate maps, error analysis diagrams and other relevant result diagrams for microseismic source location in real time, making it easier for users to intuitively understand the location results.

[0105] from Figure 5 As can be seen, the system workflow diagram in this invention includes a sensor array (16), data flow (17), underground station (18), ground control center (19), data preprocessing module (20), intelligent computing positioning module (21), result analysis and visualization module (22), and mine working face (23). Each component forms a complete closed loop through the data link.

[0106] The sensor array (16) consists of several sets of three-component detectors, which are installed on the top plate and side wall of the mine working face (23) and connected to the underground station (18) through fiber optic composite cable. The data flow (17) transmits the raw data of the sensor array (16) to the underground station (18) through full-duplex communication (transmission bandwidth ≥100Mbps). The underground station (18) has a built-in signal conditioning circuit and Beidou time synchronization module, which is responsible for the initial time stamp alignment of the signal (time stamp accuracy up to 0.1ms).

[0107] The data preprocessing module (20) receives data from the underground station (18) and completes noise suppression (signal-to-noise ratio improvement ≥15dB) and feature extraction (P / S wave arrival time marker) through an adaptive filter bank and waveform truncation unit.

[0108] The intelligent computing positioning module (21) is built on the Physical Information Neural Network (PINN) architecture. It receives the feature parameters output by the data preprocessing module (20), including the arrival time of the microseismic signal (usually the arrival time of the P-wave) signal duration and the corresponding three-component sensor spatial coordinate information, as input to the neural network. The module integrates data-driven learning and physical law constraints, and directly generates the spatial coordinates (x, y, z) and start time (t) of the seismic source at the output end, realizing accurate inversion of the seismic source.

[0109] The neural network structure employs a multi-layer fully connected feedforward network, with activation functions enhancing the network's ability to fit spatial nonlinear field characteristics. Residual connections and batch normalization techniques are also introduced to improve the stability of the training process and the model's generalization ability. The wave velocity parameter c(x,y,z) can be determined in three ways: firstly, by inverting the arrival time of the first arrival wave from the microseismic event to obtain the local equivalent wave velocity; secondly, by introducing a preset regional average wave velocity from existing geological survey results; and thirdly, by adaptively inverting the spatially correlated wave velocity field using PINN during model training, forming a weakly supervised learning structure.

[0110] During the training phase, this module uses the Adam optimizer for gradient updates, supplemented by a dynamic learning rate decay strategy to accelerate convergence. During the inference phase, it uses GPU parallel computing to quickly output four-dimensional source parameters, meeting the real-time requirements of engineering practice.

[0111] After the sensor array (16) detects a microseismic event, the raw data is transmitted to the underground station (18) via the data flow (17) to complete the time-scale alignment; the preprocessing module (20) performs data standardization processing, the positioning module (21) calls the PINN model to invert the source location, and finally the visualization module (22) generates a dynamic wavefront propagation map and stores it in the event database.

[0112] from Figure 6 As can be seen, the core sensing unit in this invention is specifically composed of the following components: a three-component high-precision probe (24), a composite explosion-proof housing (25), a calibration rotation assembly (26), an indicator light (27), an edge storage chip (28), a backup power supply (29), a Beidou time transmitter (30), and an external fiber optic composite cable (31).

[0113] The three-component high-precision probe (24) achieves multi-dimensional signal detection through an orthogonally arranged sensor array; the composite explosion-proof housing (25) provides double protection for the probe (24) - the outer stainless steel structure resists physical impact, and the inner silicone damping layer suppresses vibration interference.

[0114] The calibration rotation component (26) is linked with the three-component high-precision probe (24) to support remote control of three-axis orientation calibration; the indicator light (27) provides real-time feedback on the system status through a three-color LED array - green corresponds to normal operation, yellow indicates calibration / debugging mode, and red triggers a fault alarm.

[0115] The edge storage chip (28) is directly connected to the three-component high-precision probe (24) for temporary storage of detection data; the Beidou timer (30) works in conjunction with the edge storage chip (28) through an external fiber optic composite cable (31) to ensure that the data timestamp is synchronized with the ground system, with a time synchronization accuracy of ±0.1μs.

[0116] The backup power supply (29) automatically takes over when the power supply is interrupted, providing 72 hours of independent operation support for the probe (24), calibration components (26), etc. The external fiber optic composite cable (31) simultaneously undertakes the functions of power transmission, signal interaction and high-speed data return, forming a redundant guarantee link with the backup power supply (29).

[0117] from Figure 7 It can be seen that the intelligent control center consists of the following core units: signal receiving device (32), preprocessing chassis (33), signal indicator (34), fiber optic composite cable (35), intelligent computing module box (36), neural network computing chip matrix (37), heat dissipation hole (38), display screen (39) and display waveform (40).

[0118] Wired signal synchronization transmits the digital signal of the directly connected sensor to the preprocessing chassis (33) via fiber optic composite cable (35); wireless signal acquisition transmits 2.4GHz / 5GHz wireless signals through the antenna array of the downhole sensor array and the transmitting device (32) to cover shallow wireless signals, amplify the signals and separate the frequency bands; the signal and wireless link data are time-aligned within the preprocessing chassis (33). After synchronous sampling in the preprocessing chassis (33), the trigger signal indicator (34) switches to a blue flashing mode; the raw data is denoised and normalized and then injected into the intelligent computing module box (36).

[0119] The backplane of the intelligent computing module box (36) dynamically distributes the data stream to the computing unit and controls each group of chips (37) to execute in parallel; during the calculation process, the signal indicator (34) switches to the purple breathing light mode. The positioning data is transmitted to the display screen (39) and generated in real time; the original signal (gray) and the noise-reduced signal (blue) are displayed synchronously in the display waveform diagram (40); the touch operation command of the display end (39) is transmitted to the preprocessing chassis (33) through the reverse link (right → middle → left).

[0120] The fan speed of the heat dissipation hole (38) is dynamically adjusted according to the chip load; the temperature control data of the heat dissipation hole (38) is connected to the computing module (37) in real time, triggering the following protection mechanism: when the chip junction temperature is >75℃, the batch processing scale of the PINN network is automatically reduced; when the temperature is >85℃, the computing task migration is forcibly started. The Beidou timing signal is transmitted in reverse to the sensor end through the optical fiber composite cable (35) to calibrate the sampling clock deviation (<0.1ms).

[0121] The alarm information of the signal indicator (34) and the display screen (39) are linked: when the calculation is abnormal, the purple flashing frequency of the indicator doubles and the screen displays a gradient explosion warning; when the signal is lost, the indicator switches to red and flashes quickly, and the topology breakpoint location map pops up on the screen.

[0122] Accordingly, this embodiment provides a microseismic source localization method based on a Physical Information Neural Network (PINN). This method utilizes physical constraints to optimize the neural network training process, thereby improving the localization accuracy of microseismic events. Specifically, it includes the following steps:

[0123] Deployment of the microseismic monitoring system: The microseismic monitoring system is deployed in the mine roadways or mining area. The system includes multiple seismic sensors and data acquisition equipment, which are connected to the data acquisition server via wired or wireless means and transmit the signals to the ground data processing center.

[0124] Data Acquisition and Preprocessing: The system performs long-term continuous monitoring of microseismic events occurring within the mine, acquiring microseismic waveform data, such as... Figure 2 As shown. Data preprocessing includes denoising, filtering, normalization, etc., to extract effective waveform features.

[0125] Construction of PINN: Based on the theory of elastic wave propagation, a microseismic source location model based on PINN is constructed. This model employs a differential equation-constrained neural network training process, ensuring that the prediction results conform to the seismic wave equation.

[0126] PINN model training: The acquired microseismic data is used as input, and the spatial coordinates of the seismic source are used as the target output. The model is optimized using a physical loss function. The loss function includes a data loss term (error minimization) and a physical loss term (partial differential equation constraints).

[0127] Microseismic source location calculation: Input the newly acquired microseismic data into the trained PINN model to calculate the spatial coordinates of the source and output the source location, magnitude and uncertainty estimate.

[0128] Data verification and optimization: The positioning results are compared and analyzed with the calculation results of existing microseismic monitoring systems. Error analysis and optimization of the model are performed to improve the positioning accuracy of microseismic sources.

[0129] System Applications and Expansions: This method can be widely applied to microseismic monitoring in engineering environments such as coal mines, metal mines, and tunnels. It can also be combined with acoustic detection, ground stress monitoring, and other methods to improve the reliability of mine surrounding rock safety monitoring.

[0130] The beneficial effects of this embodiment are as follows:

[0131] Compared to traditional microseismic source location methods, this embodiment combines physical information and deep learning technology, which significantly improves the accuracy and reliability of microseismic source location, and can maintain high positioning accuracy even in complex environments and under noise interference.

[0132] The system has real-time data processing and visualization capabilities, providing users with intuitive microseismic source location results and error analysis, and supporting the needs of decision support systems.

[0133] The system in this embodiment has strong adaptability and scalability. It can be adjusted and optimized according to actual needs, does not require a large amount of labeled data, and is suitable for microseismic monitoring tasks of different scales.

[0134] This application also provides an electronic device, including: a processor, and a memory coupled to the processor, the memory being used to store a computer program; the processor being used to execute the computer program stored in the memory, so that the electronic device performs the method as described in any of the above embodiments.

[0135] Electronic devices can be computing devices such as desktop computers, laptops, handheld computers, and cloud servers. These electronic devices may include, but are not limited to, processors and memory.

[0136] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the device via various interfaces and lines.

[0137] The memory can be used to store the computer program, and the processor implements various functions of the electronic device by running or executing the computer program stored in the memory and calling the data stored in the memory.

[0138] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0139] This application also provides a storage medium, which is a computer-readable storage medium. The computer program is stored in the computer-readable storage medium, and when executed by a processor, the computer program can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0140] This application also provides a computer program product, including: a computer program or instructions that, when the computer program or instructions are run on a computer, cause the computer to perform any of the above possible implementations of the method.

[0141] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.

Claims

1. A training method for a microseismic source location model, characterized in that, The training method includes: Training data is obtained, wherein the training data is time-related microseismic waveform data obtained by sensor sensing; and The microseismic source location model is trained using the training data. The microseismic source location model adopts a physical information neural network (PINN). The PINN takes the microseismic waveform data as input and the spatial coordinates of the source and the time of occurrence as output. The overall loss function used by PINN includes a data loss term and a physical loss term, wherein the physical loss term is defined by a wave equation; The wave velocity parameter in the wave equation is determined by one of the following methods: obtaining the equivalent wave velocity through first arrival wave travel time inversion, the average wave velocity of a preset geological region, and the spatially correlated wave velocity field obtained by neural network inversion. The data loss term is defined by the predicted source displacement and the actual measured displacement; The overall loss function employs an adaptive weighting strategy; The overall loss function L total for: Among them, L data For the data loss term, L phys Let be the physical loss term, α be the weight of the data loss term, λ be the weight of the physical loss term, i be the data point index, N be the total number of sampling points, j be the physical constraint index, and M be the total number of physical constraint points. For the predicted medium response displacement, This represents the actual measured displacement of the medium. Let i be the spatial location of the i-th data point. Let i be the sampling time for the i-th data point. Let j be the spatial location of the j-th data point. Let c be the sampling time of the j-th data point, and c be the wave velocity. is the Laplace operator, representing the change in wave propagation in space, and k is the number of training iterations.

2. The training method as described in claim 1, characterized in that, The training method further includes: batch normalizing the microseismic waveform data, wherein the batch normalization specifically includes: Based on the time-frequency analysis characteristics of the microseismic waveform data, the sliding window type is determined; and When the time-frequency analysis characteristics of the microseismic waveform data indicate that the microseismic waveform data is a high-frequency transient signal, a short window is used for batch normalization; when the time-frequency analysis characteristics of the microseismic waveform data indicate that the microseismic waveform data is a low-frequency continuous signal, a long window is used for batch normalization.

3. A method for intelligent localization of microseismic sources, characterized in that, The positioning method includes: Obtain the current microseismic waveform data; and The current microseismic waveform data is input into the microseismic source localization model trained by the training method described in any one of claims 1-2, and the current spatial coordinates of the source and the current time of occurrence are obtained as output.

4. A microseismic source intelligent positioning system, characterized in that, The system includes: The sensor module is used to obtain the raw data of the current micro-vibration waveform; The preprocessing module is used to standardize the raw data of the current microseismic waveform to obtain the current microseismic waveform data. The intelligent calculation and positioning module is used to call the microseismic source positioning model trained by the training method described in any one of claims 1-2 to process the current microseismic waveform data, and obtain the current spatial coordinates of the source and the time of occurrence; and The visualization module is used to generate and store dynamic wavefront propagation diagrams based on the current spatial coordinates of the seismic source.

5. The system as described in claim 4, characterized in that, The sensor module includes: A three-component high-precision probe using an orthogonally arranged sensor array; The calibrated rotating component is linked with the three-component high-precision probe to support remote control of triaxial orientation calibration. An edge storage chip is directly connected to the three-component high-precision probe for temporary storage of monitoring data; and The BeiDou timer is connected to the edge storage chip via an optical fiber composite cable to ensure that the data timestamp is synchronized with the ground system.

Images