Multi-source earthquake caving area positioning device and method for underground roadway mining

Abstract

The invention belongs to the field of coal mine roadway disaster monitoring, and particularly discloses a multi-source earthquake caving area positioning device and method for underground roadway mining, and the device comprises a plurality of multi-source seismic source vehicles which are used for transmitting pulse signals; the three-dimensional detector array is used for collecting waveforms generated by a micro-seismic event in real time; the edge acquisition and clock synchronization unit is used for processing the pulse signal and the micro-seismic waveform and then outputting the pulse signal and the micro-seismic waveform; the ground data processing center is used for receiving the processed micro-seismic waveform, performing micro-seismic event density calculation and judgment according to the processed micro-seismic waveform, and calling a deep reinforcement learning scheduling engine to issue an excitation sequence to a corresponding excitation module according to a judgment result; and then carrying out adaptive grid chromatography and multi-scale finite difference full waveform inversion on the new pulse signal generated by the excitation module and the synchronously generated micro-seismic waveform to realize the space imaging of the caving area. According to the invention, the space imaging of the caving area can be realized under the condition of non-stop production.

Description

Technical Field

[0001] This application belongs to the field of coal mine roadway disaster monitoring, and more specifically, relates to a multi-source seismic collapse zone location device and method for underground roadway mining. Background Technology

[0002] In the field of coal mine roadway disaster monitoring, the core objective of using multi-source seismic signals to perform three-dimensional localization of roof collapses in coal mining goafs is to provide real-time, detailed, and visualized spatial imaging methods for identifying potential roof collapse hazards under continuous underground production conditions.

[0003] Currently, microseismic monitoring systems are commonly used for assessing the stability of underground roofs. These systems typically employ a three-component geophone network installed in the surrounding rock of the tunnel and deep boreholes. Through A / D sampling and time-difference inversion, natural microseismic events are located in three dimensions. Typical products include commercial systems from ESG (Canada) and IMS (Germany). A common characteristic of these systems is that they rely on the wave energy released when the rock mass fractures to acquire signals, thus allowing location analysis only after an incident has occurred or a fracture has formed. When event density is low or the frequency of occurrence is unstable, the monitoring network may experience location blind spots and cannot actively image the geometry of potential collapses.

[0004] To address the need for advanced detection, some research and patents have introduced active seismic methods. Patent CN112285802B proposes a combined "tunnel seismic-transient electromagnetic" technology, first using tunnel reflection seismic data to outline a velocity model, and then using this model to constrain transient electromagnetic inversion, aiming to improve the resolution of anomalies ahead. Patent CN109459787A utilizes a channel wave full-waveform inversion method to image fine coal seam structures, relying on Love waves excited by boreholes on one side of the wall to obtain high-resolution data. While these approaches have achieved results in identifying small structures or predicting advanced obstacles, they generally rely on a single active seismic source (such as a small-charge explosive or shockwave), requiring drilling to be stopped and limited by the number of blasts; simultaneously, the computational load of algorithms such as full-waveform inversion is enormous, hindering real-time field applications. Small-bore excitation technologies such as TSP (Tunnel Seismic Prediction) also face similar limitations: fewer excitations, narrow coverage angle, and limited resolution.

[0005] In recent years, multi-parameter fusion early warning platforms have added sensors such as anchor stress and support resistance to the microseismic data, outputting a hazard index through empirical thresholds or machine learning models. However, their focus remains on risk warning rather than the three-dimensional morphology of the collapse zone. Due to the lack of active wavefield information, these systems struggle to provide intuitive spatial imaging before a large number of microseismic events have occurred.

[0006] Therefore, how to achieve spatial imaging of the collapse zone without interrupting production is an urgent problem to be solved. Summary of the Invention

[0007] In view of the shortcomings of the prior art, the purpose of this application is to provide a multi-source seismic collapse zone positioning device and method for underground roadway mining, which can realize spatial imaging of the collapse zone without stopping production.

[0008] To achieve the above objectives, in a first aspect, this application provides a multi-source seismic collapse zone location device for underground roadway mining, comprising: Multiple multi-source seismic source vehicles are set up in coal mine roadways. They integrate three pluggable excitation modules: electromagnetic impact seismic source, water hole electric spark seismic source, and micro-charge directional blasting seismic source. These modules are used to receive and control the corresponding excitation modules to transmit pulse signals with known time and waveform to the surrounding roadway according to the excitation sequence. The three-dimensional geophone array includes a MEMS vector geophone anchored to the tunnel wall, a broadband three-component seismometer buried in deep holes in the roof and floor plates, and an accelerometer module fixed by a hydraulic support, used to acquire waveforms generated by microseismic events in real time. The edge acquisition and clock synchronization unit is used to acquire and process pulse signals and micro-vibration waveforms before outputting them. This processing includes timestamp calibration. The ground data processing center receives continuous microseismic waveforms uploaded by the edge acquisition and clock synchronization units; it preprocesses the waveforms and extracts microseismic events, calculates the microseismic event density within a set time window and compares it with a threshold; when the density exceeds the threshold or is manually triggered, it calls the deep reinforcement learning scheduling engine to generate a source excitation sequence and sends it to the corresponding excitation module; subsequently, it receives the pulse signal generated by the excitation module and the real-time synchronized microseismic waveforms, and sends both to adaptive grid tomography to obtain the instantaneous velocity field. Using this velocity field as a priori input, it performs multi-scale finite difference full waveform inversion to obtain the three-dimensional impedance anomaly of the collapse zone, realizing spatial imaging of the collapse zone and outputting it to visualization and database.

[0009] The beneficial effects of the multi-source seismic collapse zone location device for underground roadway mining provided in this application are as follows: First, the three-dimensional geophone array operates normally, continuously acquiring roof micro-vibrations, thereby updating the event density in real time without interrupting production. When the density index increases, indicating an intensified collapse hazard, the deep reinforcement learning scheduling engine selects only seismic source schemes allowed within the constraints of gas level, noise, and imaging depth, such as low-energy electromagnetic shock or directional micro-explosive blasting. The amount of explosive used in each excitation is ≤80 g TNT, and the directional shield faces the roadway wall, which complies with the "local blasting without interrupting production" clause in the "Safety Regulations for Coal Mines" for underground mining, thus eliminating the need for evacuation. The excitation-reception link is synchronized at the 10 µs level by an edge time synchronization closed loop, allowing the active waveform and synchronous micro-vibrations to be inverted in the same model. Then, the GPU-FWI is used to fuse the initial velocity model provided by the passive micro-vibrations with the high-frequency details of the active wave field, outputting a 20 m × 20 m × 10 m three-dimensional impedance map in only about 120 s. The entire process does not affect the continuous advancement of the tunneling machine, nor does it require stopping ventilation, power supply, or support operations. Therefore, it essentially achieves the spatial imaging capability of the collapse zone through "real-time scanning-imaging-feedback without production interruption".

[0010] As a further preferred embodiment, the pulse signal experiences travel time delay, amplitude attenuation, and spectral distortion when it encounters loose collapsed material, fracture zone, or intact surrounding rock during propagation in the tunnel. The microseismic waveform refers to the microfractures generated in the roof rock mass when it is subjected to mining or stress concentration, which will radiate P-waves, S-waves and high-frequency shear fracture waves of 50Hz to 5kHz.

[0011] As a further preferred option, two seismic source vehicles are used to travel independently on both sides of the same working face.

[0012] As a further preferred embodiment, the main frequency range of the electromagnetic shock source is 20-200Hz; the upper limit of the frequency band of the water-hole electric spark source is 2kHz; the charge of the micro-charge directional blasting source is 30-80g TNT equivalent and is equipped with a steel shielding shell, and the blasting gas is guided to the non-working side through a coaxial exhaust pipe.

[0013] As a further preferred embodiment, the multi-source seismic source vehicle has a tracked structure and is certified for both explosion-proof and intrinsically safe operation. The top of the vehicle integrates an intrinsically safe power supply, a hydraulic servo cylinder, and an interlock controller, which can stop every 5 to 8 meters along the sidewall of the tunnel and automatically adjust the preload of the coupling rod to 0.8 MPa. As a further preferred embodiment, the pulse signal generated by the seismic source and the microseismic waveform signal acquired by the three-dimensional detector array are fed into the edge acquisition and clock synchronization unit through the same intrinsically safe gigabit single-mode optical fiber.

[0014] As a further preferred embodiment, the edge acquisition and clock synchronization unit adopts an FPGA-SoC architecture, combined with a GNSS disciplined rubidium clock and a fiber optic PTP dual-redundant timing system. The low-frequency channel is sampled at 24 bits and 2kSps, and the high-frequency channel is sampled at 32 bits and 10kSps. Half-wavelet packet filtering, STA / LTA composite triggering and data compression are performed on the chip.

[0015] As a further preferred embodiment, the ground data center employs four GPU servers and two FPGA acceleration cards, with a dual-redundant fiber optic ring network upload bandwidth of no less than 10Gbps.

[0016] Secondly, this application provides a method for locating multi-source seismic caving zones based on any of the aforementioned devices, comprising the following steps: S10: Acquire microseismic waveforms continuously acquired by the three-dimensional geophone array, calculate the microseismic event density based on the microseismic waveforms and compare it with a threshold. S20: When the density of microseismic events within the set time window exceeds the threshold or is manually triggered, the deep reinforcement learning engine is invoked to generate a seismic source excitation sequence and send it to the corresponding excitation module. S30, acquire the pulse signal generated by the excitation module and the micro-vibration waveform generated in real time synchronously; S40, the pulse signal and microseismic waveform from step S30 are jointly fed into adaptive grid tomography to obtain the instantaneous velocity field, and the velocity field is used as the prior input for multi-scale finite difference full waveform inversion to obtain the three-dimensional impedance anomaly of the collapse zone, thereby realizing the spatial imaging of the collapse zone and outputting it to visualization and database.

[0017] As a further preferred embodiment, the decision-making criteria of the deep reinforcement learning engine include: Safety constraints: If the gas concentration is higher than 1.0%, select one of the electromagnetic shock source and the water hole electric spark source to send an excitation sequence; Imaging depth: When penetration of at least 50 m is required, use a micro-charge directional blasting source for blasting; when the depth is no more than 30 m and resolution is important, send an excitation sequence to an electromagnetic shock source. Noise environment: When the tunneling machine causes noise to spike in the range of 100 to 300 Hz, select a water hole electric spark source with a main frequency of 20 to 80 Hz to send out the excitation sequence in order to avoid the noise band.

[0018] It is understandable that the beneficial effects of the second aspect mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0019] Figure 1This is an overall structural block diagram of the multi-source seismic collapse zone positioning device for underground roadway mining provided in the embodiments of this application; Figure 2 This is a schematic diagram of the multi-source seismic source vehicle and its deployment provided in the embodiments of this application; Figure 3 This is a flowchart of the active-passive joint imaging data flow and algorithm provided in the embodiments of this application. Detailed Implementation

[0020] 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.

[0021] It should be understood that, in the description of this application, the term "multiple" means two or more, unless otherwise expressly and specifically defined.

[0022] like Figure 1 As shown, this application provides a multi-source seismic collapse zone location device for underground roadway mining. The device mainly includes multiple multi-source seismic source vehicles, a three-dimensional geophone array, an edge acquisition and clock synchronization unit, and a ground data processing center.

[0023] The multi-source seismic source vehicle provided in this application is installed in a coal mine roadway and integrates three pluggable excitation modules: an electromagnetic impact seismic source, a water-hole electric spark seismic source, and a micro-charge directional blasting seismic source. It is used to receive and control the corresponding excitation module to transmit pulse signals with known time and known waveform to the surrounding roadway according to the excitation sequence.

[0024] In one embodiment, such as Figure 2 As shown, the multi-source seismic source vehicle can be a tracked underground multi-source seismic source vehicle with a width of 0.8m and certified for both explosion-proof and intrinsically safe operation. The vehicle's interior has standardized interfaces for three pluggable excitation modules: an electromagnetic impact seismic source capable of releasing 20–2000J of kinetic energy within 0.2s, covering a main frequency range of 20–200Hz; a water-hole electric spark seismic source placed in a pre-drilled φ42mm deep hole, with a single discharge energy of 2kJ and a frequency band upper limit of 2kHz; and a micro-charge directional blasting seismic source loaded with 30–80g of TNT equivalent and equipped with a steel shielding shell, guiding the blasting gas to the non-working side via a coaxial exhaust pipe. The seismic source vehicle's top integrates an intrinsically safe power supply, a hydraulic servo cylinder, and an interlock controller, allowing it to stop every 5–8m along the roadway sidewall and automatically adjust the coupling rod preload to 0.8MPa. The excitation trigger pulse from the seismic source and the data collected by the detector are transmitted in a closed loop through the same intrinsically safe high-speed CAN-fiber hybrid bus, ensuring that the excitation control command and the first wave response are synchronized within the same time reference.

[0025] This embodiment modularly integrates three types of seismic sources—electromagnetic shock, water-hole electric spark, and micro-charge directional blasting—on a tracked chassis with dual explosion-proof and intrinsically safe certifications. It achieves second-level switching and safety interlocking through quick-connect interfaces and interlock controllers, covering the 20Hz–2kHz frequency band and providing 20J–2kJ continuously adjustable excitation energy.

[0026] The three-dimensional geophone array provided in this application includes a MEMS vector geophone anchored to the tunnel wall, a broadband three-component seismometer buried in deep holes in the top and bottom plates, and an acceleration module fixed by a hydraulic support, used to acquire waveforms generated by microseismic events in real time.

[0027] In one embodiment, the three-dimensional detector array consists of three channels. MEMS vector detectors with a sensitivity of 4000 are resin-anchored to the tunnel wall at 5m intervals. The frequency band is 10–1000 Hz, responsible for capturing high-frequency reflected signals. A 6m deep borehole is drilled in the direction of the top and bottom plates, and a broadband three-component seismometer is installed at the bottom of the borehole, operating in the frequency band of 1–500 Hz with an equivalent noise level of 0.25. This creates a three-dimensional coverage and keeps the area away from roadway disturbances. To synchronously record ultra-high frequency information of roof deformation, an integrated acceleration module is welded to the hydraulic support column every 10m, which can sense transient impacts in the range of 0-5kHz.

[0028] This embodiment of the three-dimensional detector array consists of a multi-scale receiving network composed of a tunnel wall MEMS vector detector, a high- and low-aperture three-component broadband seismometer, and a support accelerometer module. It can achieve real-time high dynamic range acquisition of signals across the entire frequency band from 10Hz to 5kHz and obtain complete wavefield polarization information using a three-dimensional deployment. The edge acquisition and clock synchronization unit provided in this application is used to acquire and process pulse signals and micro-vibration waveforms before outputting them. This processing includes timestamp calibration.

[0029] In one embodiment, all channel signals can be fed into the edge acquisition and clock synchronization unit via intrinsically safe gigabit single-mode fiber. This unit adopts an FPGA-SoC architecture, with low-frequency channels sampled at 24 bits and 2kSps, and high-frequency channels sampled at 32 bits and 10kSps. Half-wavelet packet filtering, STA / LTA composite triggering, and data compression are performed on-chip. System timing is transmitted downlink from a ground-based GNSS disciplined rubidium clock, distributed to each node via fiber optic PTP, and maintained by a temperature-compensated crystal oscillator, keeping the overall network synchronization error within 10µs.

[0030] In this embodiment, the FPGA-SoC edge node adopts a 24 / 32 bit hybrid resolution sampling architecture, combined with a GNSS disciplined rubidium clock + fiber optic PTP dual redundant timing system, which can ensure that the time error of the entire network is ≤10 µs; the excitation trigger pulse and the first wave response are transmitted in a closed loop through the same intrinsically safe CAN-fiber optic bus to achieve full-chain synchronization of excitation-reception.

[0031] The ground data processing center provided in this application is used to receive continuous microseismic waveforms uploaded by the edge acquisition and clock synchronization unit; preprocess the waveforms and extract microseismic events, calculate the microseismic event density within a set time window and compare it with a threshold; when the density exceeds the threshold or is manually triggered, the deep reinforcement learning scheduling engine is invoked to generate a source excitation sequence and send it to the corresponding excitation module; subsequently, the pulse signal generated by the excitation module and the real-time synchronized microseismic waveform are received, and the two are jointly sent to adaptive grid tomography to obtain the instantaneous velocity field, and the velocity field is used as the prior input for multi-scale finite difference full waveform inversion to obtain the three-dimensional impedance anomaly of the collapse zone, realize the spatial imaging of the collapse zone and output it to visualization and database.

[0032] In one embodiment, the ground data processing center can employ four GPU servers and two FPGA acceleration cards, with a dual-redundant fiber optic ring network ensuring an upload bandwidth of no less than 10Gbps. With this configuration, the system can maintain a network clock synchronization error of no more than 10µs for extended periods while operating underground, achieving a single active scan imaging coverage volume of 20m×20m×10m and completing reconstruction within 120 seconds.

[0033] Specifically, the data processing section of the data processing center provided in this application can run in a loop under a Linux environment, following the sequence of "passive monitoring—active scanning—joint inversion—3D visualization." For example... Figure 3The system first performs wavelet packet energy spectrum analysis on the real-time stream (referring to the raw data stream of continuous waveforms in the array) and extracts microseismic events through K-means spatiotemporal clustering. If the event density index exceeds 0.6 within 30 minutes (this index is the normalized value of the frequency of microseismic triggering per unit time and unit volume, taking the range of 0-1) or is manually triggered by the scheduler, the deep reinforcement learning scheduling engine is invoked to generate the optimal excitation sequence (source type, energy level, and azimuth) based on the historical event distribution, remaining blasting permits, and spatial obstacles. After the source vehicle is sequentially excited, the edge acquisition unit immediately marks the absolute time and uploads the first wave pickup; the ground center performs adaptive grid tomography inversion by combining the microseismic travel time with the first wave travel time of the active source. After all field excitations are completed and the data quality is confirmed, the data processing center immediately calls the wavefield joint inversion module to immediately enter the wavefield joint inversion stage. The system updates the 3D P-wave velocity field in real time and uses it as a priori input to a multi-scale finite-difference full-waveform inversion module. With parallel acceleration using a 96k CUDA core, impedance imaging of a 20m×20m×10m grid is completed within 120 seconds. The obtained low-velocity, high-attenuation anomaly is probabilistically segmented using a convolutional neural network to generate a 3D probabilistic volume of the collapse zone. This volume is then rendered in volumetric form on the control room's large screen and AR headset using a self-developed WebGL kernel, enabling real-time, precise, and visual localization of the mining collapse. The results are simultaneously written to a PostgreSQL database for tracking, comparison, and safety decision-making. The system then reverts to a passive monitoring loop. If the algorithm detects hardware anomalies, missing data, or insufficient computing resources, the control logic automatically switches to a pure passive monitoring mode and sends an audible and visual alarm to the dispatch console, ensuring hardware safety and data continuity.

[0034] The working principle of the multi-source seismic collapse zone location device provided in this application is as follows: After the seismic source vehicle stops along the sidewall of the tunnel, it selects an electromagnetic shock, water-hole electric spark, or micro-charge directional blasting module based on the excitation sequence sent from the ground and instantaneously couples it to the surrounding rock. The array collects microseismic and excitation waveforms in real time. After 24 / 32-bit synchronous sampling, preprocessing, and timestamp calibration by the edge unit, the data is uploaded through a dual-redundant fiber optic ring network. The ground part includes a GPU-FPGA data center, a visualization terminal, and a database: the uploaded data first enters the passive monitoring thread to complete event triggering and location. If the event density index exceeds the threshold or manual intervention occurs within 30 minutes, the deep reinforcement learning scheduling engine automatically generates a new excitation list and sends it back downhole. Subsequently, the same batch of data is sent to the joint tomography and full waveform inversion module to generate a real-time velocity-impedance model. After probability segmentation, the results are pushed to the dispatch room and AR helmet in 3D volume rendering form and written to the database to form a closed loop. Then, the system status is reset to passive monitoring.

[0035] The multi-source seismic collapse zone location device provided in this application integrates passive microseismic sources and multiple types of active seismic sources. The multi-source seismic source vehicle (electromagnetic / electro-spark / micro-explosive blasting) is the active wavefield generation unit. Its main value lies in clarifying the roof rock mass: each excitation emits a pulsed wavefield with a known time and waveform to the surrounding tunnel; these pulses, when encountering loose collapse bodies, fracture zones, or intact surrounding rock during propagation, produce significant calculable travel time delays, amplitude attenuation, and spectral distortion. The ground server, during inversion, utilizes this forward modeling relationship of "known source—unknown medium—known receiver" to constrain the velocity and impedance fields, thereby supplementing the "static structure" that passive microseismic sources cannot display. A three-dimensional geophone array (wall-mounted MEMS + roof and floor borehole seismometers + support accelerometers) is responsible for recording the passive microseismic waveforms. The "microseismic event" here refers to the microfractures generated in the roof rock mass during mining or stress concentration, which radiate P-waves, S-waves, and high-frequency shear fracture waves of 50 Hz–5 kHz. The array acquires the polarization direction and arrival time difference of the waves through a three-dimensional deployment. The server first uses these to perform event localization and real-time velocity modeling. Then, the actively excited waveforms received by the same array and the microseismic travel times are sent together to "joint tomography + full waveform inversion," thereby simultaneously imaging the "evolving dynamic source" and the "existing structural anomaly" in an integrated grid, achieving a comprehensive identification of the location, extent, and evolution stage of the collapse zone.

[0036] The beneficial effects of the multi-source seismic collapse zone location device for underground roadway mining provided in this application are as follows: First, the three-dimensional geophone array operates normally, continuously acquiring roof micro-vibrations, thereby updating the event density in real time without interrupting production. When the density index increases, indicating an intensified collapse hazard, the deep reinforcement learning scheduling engine selects only seismic source schemes allowed within the constraints of gas level, noise, and imaging depth, such as low-energy electromagnetic shock or directional micro-explosive blasting. The amount of explosive used in each excitation is ≤80 g TNT, and the directional shield faces the roadway wall, which complies with the "local blasting without interrupting production" clause in the "Safety Regulations for Coal Mines" for underground mining, thus eliminating the need for evacuation. The excitation-reception link is synchronized at the 10 µs level by an edge time synchronization closed loop, allowing the active waveform and synchronous micro-vibrations to be inverted in the same model. Then, the GPU-FWI is used to fuse the initial velocity model provided by the passive micro-vibrations with the high-frequency details of the active wave field, outputting a 20 m × 20 m × 10 m three-dimensional impedance map in only about 120 s. The entire process does not affect the continuous advancement of the tunneling machine, nor does it require stopping ventilation, power supply, or support operations. Therefore, it essentially achieves the spatial imaging capability of the collapse zone through "real-time scanning-imaging-feedback without production interruption".

[0037] In one embodiment, two source vehicles can be configured to operate independently on opposite sides of the same working face. The data processing center assigns "vehicle ID + module + excitation time" based on the optimal excitation sequence and sends this information via fiber optic bus. Upon receiving the "target time T," the interlock controller inside each source vehicle self-calibrates its clock and excites the vehicle within ±1 ms to ensure multi-vehicle coordination.

[0038] In one embodiment, the ground data processing center provided in this application can adopt a five-ring data processing logic of "data-criteria-control-imaging-feedback": The continuous waveform stream from the well is used as the raw input, and microseismic events are extracted in real time from the energy spectrum-clustering unit accelerated by CUDA, and the event density is calculated; this density value, together with the historical trends in the database, constitutes a reinforcement learning state vector, and the scheduling network outputs the next batch of source types, energy, and timing sequences, which are immediately transmitted via optical fiber; the newly generated waveforms and synchronous microseismic travel times are sent to adaptive grid tomography to obtain the instantaneous velocity field, and then enter multi-scale finite difference FWI to refine the impedance; the obtained three-dimensional anomaly volume is generated into a probability volume by a convolutional network, and then pushed to the WebGL rendering service for visualization, triggering API write-back to the database and security interlocking logic. If any link (acquisition, network, GPU resources) malfunctions, the daemon process will switch to a purely passive monitoring mode and alarm the scheduling terminal, thereby ensuring that the causal relationship and fault-tolerant closed loop between the algorithm, data stream, and hardware actions are clear and controllable.

[0039] This embodiment provides rapid visualization of 3D probabilistic volumes: the GPU-accelerated WebG rendering kernel can convert impedance anomalies into volumetric probability clouds of collapse zones within <2 seconds, and simultaneously push them to the large screen in the dispatch room and the AR helmet, enabling collaborative decision-making between the underground and surface. Furthermore, this embodiment provides safety and fault tolerance, as well as mode switching: the hardware, self-test, and data link monitoring modules automatically switch to a purely passive monitoring mode and trigger audible and visual alarms when an anomaly is detected, ensuring continuous observation and data integrity even in the event of seismic source failure, power fluctuations, or network interruptions.

[0040] In one embodiment, the selection logic for the three types of seismic sources is automatically completed by the server-side "deep reinforcement learning scheduling engine". The decision criteria include: safety constraints: if the gas concentration is higher than 1.0%, micro-explosive blasting is prohibited, and only "electromagnetic shock + electric spark" is selected; imaging depth: when penetration ≥ 50 m is required, micro-explosive blasting with the highest energy is given priority; when the depth is ≤ 30 m and resolution is important, high-frequency electromagnetic shock is used; noise environment: when the tunneling machine operation causes noise to spike in 100–300 Hz, water hole electric spark with a main frequency of 20–80 Hz is selected to avoid the noise zone.

[0041] The algorithm updates the strategy in real time based on the distribution of microseismic events in the last 30 minutes, historical inversion errors, and measured gas values, and outputs a quadruple of "source type + energy + orientation + excitation time". Therefore, the underground personnel do not need to make manual decisions, but only need to ensure that the seismic source vehicle is on standby at the designated parking point.

[0042] The deep reinforcement learning scheduling engine provided in this embodiment constructs a state space based on the spatial distribution of microseismic events, blasting permits, and roadway obstacle conditions. It uses a deep Q-network to output the optimal sequence of source type, energy level, and azimuth angle. Under the premise of ensuring safety, it obtains the maximum imaging gain with the fewest excitations. Field tests have reduced the scanning time by 45%.

[0043] Based on the same inventive concept, this application also provides a method for locating multi-source seismic collapse zones in underground roadway mining, comprising the following steps: Step S10: Acquire microseismic waveforms continuously acquired by the three-dimensional detector array, calculate the microseismic event density based on the microseismic waveforms and compare it with a threshold. Step S20: When the density of microseismic events within the set time window exceeds the threshold or is manually triggered, the deep reinforcement learning engine is invoked to generate a seismic source excitation sequence and send it to the corresponding excitation module. Step S30: Obtain the pulse signal generated by the excitation module and the micro-vibration waveform generated in real time; In step S40, the pulse signal and microseismic waveform from step S30 are jointly fed into adaptive grid tomography to obtain the instantaneous velocity field. This velocity field is then used as a priori input for multi-scale finite difference full waveform inversion to obtain the three-dimensional impedance anomaly of the collapse zone, thereby realizing spatial imaging of the collapse zone and outputting it to visualization and database.

[0044] It should be noted that the detailed implementation methods of each step provided in this application can be found in the detailed description of the foregoing device embodiments, and will not be repeated here.

[0045] Compared with existing technologies, the advantages of the multi-source seismic collapse zone location device and method provided in this application are: (1) After implementing this application, spatial imaging of the goaf and collapse area can be completed in real time without interrupting tunneling production: based on theoretical models and on-site parameter calculations, the linear positioning error of the collapse boundary within approximately 50m in front of the work face can be controlled to around 6m, while the common error of the traditional "micro-seismic + TSP" mode is 8–12m, resulting in an equivalent accuracy improvement of approximately 40%. The increased coverage angle brought about by multi-source excitation reduces the monitoring blind zone volume to less than 10%, significantly enhancing the integrity of the roadway roof stability assessment. Reinforcement learning scheduling reduces unnecessary excitation, compressing the single scan time from the conventional 40–45min to 25–30min, ensuring that safety detection and production rhythm can proceed in parallel.

[0046] (2) In terms of safety and environmental protection, directional micro-charge blasting and electromagnetic shock replace large-charge explosives, and the annual consumption of pyrotechnics is expected to be no more than one-third of the original system; the ignition process is carried out under intrinsically safe interlock protection, which reduces the risk of gas explosion, while also reducing roadway vibration and dust disturbance, making it more friendly to the underground working environment. Edge filtering and data compression reduce the original data flow of the underground-to-surface link by about half, thereby reducing the power consumption of fiber optic transmission and the overhead of surface storage.

[0047] (3) The operation and maintenance process has also been optimized: the standardized plug-in seismic source and integrated CAN-fiber optic bus have reduced the daily deployment and maintenance workload from "two shifts of five people" to "one shift of two people", and the labor intensity and labor cost are expected to decrease by more than 30%. The fully automatic waveform acquisition, grid adaptation and one-click visualization at the software level allow dispatchers to obtain the three-dimensional probabilistic volume without participating in tedious data preprocessing, providing an intuitive basis for roof management and support parameter adjustment. Overall, the present invention shows significant comprehensive advantages over the existing technology in terms of accuracy, efficiency, safety, energy consumption and labor input, providing a more economical, efficient and easy-to-use technical approach for the active prevention and control of underground roof disasters.

[0048] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A multi-source seismic collapse zone positioning device for underground roadway mining, characterized in that, include: Multiple multi-source seismic source vehicles are set up in coal mine roadways. They integrate three pluggable excitation modules: electromagnetic impact seismic source, water hole electric spark seismic source, and micro-charge directional blasting seismic source. These modules are used to receive and control the corresponding excitation modules to transmit pulse signals with known time and waveform to the surrounding roadway according to the excitation sequence. The three-dimensional geophone array includes a MEMS vector geophone anchored to the tunnel wall, a broadband three-component seismometer buried in deep holes in the roof and floor plates, and an accelerometer module fixed by a hydraulic support, used to acquire waveforms generated by microseismic events in real time. The edge acquisition and clock synchronization unit is used to acquire and process pulse signals and micro-vibration waveforms before outputting them. This processing includes timestamp calibration. The ground data processing center is used to receive continuous microseismic waveforms uploaded by the edge acquisition and clock synchronization unit; preprocess the waveforms and extract microseismic events; calculate the microseismic event density within a set time window and compare it with a threshold. When the density exceeds the threshold or is manually triggered, the deep reinforcement learning scheduling engine is invoked to generate a seismic source excitation sequence and send it to the corresponding excitation module. Then, the pulse signal generated by the excitation module and the real-time synchronized microseismic waveform are received, and the two are jointly sent to adaptive grid tomography to obtain the instantaneous velocity field. The velocity field is then used as the prior input for multi-scale finite difference full waveform inversion to obtain the three-dimensional impedance anomaly of the collapse zone, realizing the spatial imaging of the collapse zone and outputting it to visualization and database.

2. The multi-source seismic collapse zone positioning device for underground roadway mining as described in claim 1, characterized in that, When the pulse signal encounters loose collapsed material, fracture zone or intact surrounding rock during its propagation in the tunnel, it will experience travel time delay, amplitude attenuation and spectral distortion. The microseismic waveform refers to the microfractures generated in the roof rock mass when it is subjected to mining or stress concentration, which will radiate P-waves, S-waves and high-frequency shear fracture waves of 50 Hz to 5 kHz.

3. The multi-source seismic collapse zone positioning device for underground roadway mining as described in claim 1, characterized in that, Two seismic source vehicles were used to travel independently on opposite sides of the same working face.

4. The multi-source seismic collapse zone positioning device for underground roadway mining as described in claim 1, characterized in that, The electromagnetic shock source has a main frequency range of 20-200Hz; the water-hole electric spark source has a frequency band upper limit of 2kHz; the micro-charge directional blasting source has a charge of 30-80g TNT equivalent and is equipped with a steel shielding shell, and the blasting gas is guided to the non-working side through a coaxial exhaust pipe.

5. The multi-source seismic collapse zone positioning device for underground roadway mining as described in claim 1, characterized in that, The multi-source seismic source vehicle is a tracked structure with explosion-proof and intrinsically safe dual certifications. The top of the vehicle integrates an intrinsically safe power supply, a hydraulic servo cylinder, and an interlock controller. It can stop every 5 to 8 meters along the side wall of the tunnel and automatically adjust the preload of the coupling rod to 0.8 MPa.

6. The multi-source seismic collapse zone positioning device for underground roadway mining as described in claim 1, characterized in that, The pulse signal generated by the seismic source and the micro-seismic waveform signal acquired by the three-dimensional detector array are fed into the edge acquisition and clock synchronization unit through the same intrinsically safe gigabit single-mode optical fiber.

7. The multi-source seismic collapse zone positioning device for underground roadway mining as described in claim 1, characterized in that, The edge acquisition and clock synchronization unit adopts an FPGA-SoC architecture, combined with a GNSS disciplined rubidium clock and fiber optic PTP dual-redundant timing system. The low-frequency channel is sampled at 24 bits and 2kSps, and the high-frequency channel is sampled at 32 bits and 10kSps. Half-wavelet packet filtering, STA / LTA composite triggering and data compression are completed on the chip.

8. The multi-source seismic collapse zone positioning device for underground roadway mining as described in claim 1, characterized in that, The ground data center uses four GPU servers and two FPGA acceleration cards, with a dual-redundant fiber optic ring network upload bandwidth of no less than 10Gbps.

9. A method for locating multi-source seismic collapse zones based on the device described in any one of claims 1 to 8, characterized in that, Includes the following steps: S10: Acquire microseismic waveforms continuously acquired by the three-dimensional geophone array, calculate the microseismic event density based on the microseismic waveforms and compare it with a threshold. S20: When the density of microseismic events within the set time window exceeds the threshold or is manually triggered, the deep reinforcement learning engine is invoked to generate a seismic source excitation sequence and send it to the corresponding excitation module. S30, acquire the pulse signal generated by the excitation module and the micro-vibration waveform generated in real time synchronously; S40, the pulse signal and microseismic waveform from step S30 are jointly fed into adaptive grid tomography to obtain the instantaneous velocity field, and the velocity field is used as the prior input for multi-scale finite difference full waveform inversion to obtain the three-dimensional impedance anomaly of the collapse zone, thereby realizing the spatial imaging of the collapse zone and outputting it to visualization and database.

10. The method for locating multi-source seismic collapse zones as described in claim 9, characterized in that, In step S20, the decision-making criteria of the deep reinforcement learning engine include: Safety constraints: If the gas concentration is higher than 1.0%, select one of the electromagnetic shock source and the water hole electric spark source to send an excitation sequence; Imaging depth: When penetration of at least 50 m is required, use a micro-charge directional blasting source for priority; when the depth is no more than 30 m and resolution is important, send an excitation sequence to an electromagnetic shock source. Noise environment: When the tunneling machine causes noise to spike in the range of 100 to 300 Hz, select a water hole electric spark source with a main frequency of 20 to 80 Hz to send out the excitation sequence in order to avoid the noise band.

Images