Method for diagnosing loosening of mine structure

Abstract

The application provides a mine structure loosening diagnosis method, which comprises the following steps: acquiring multi-source monitoring data and ventilation record data in a ventilation flow field, wherein the multi-source monitoring data comprises wind speed monitoring data, sound disturbance data and optical fiber sensing data; preprocessing the multi-source monitoring data, and fusing the multi-source monitoring data and the ventilation record data into fused monitoring data; calculating the abnormality degree, trend score and consistency index of the fused monitoring data; performing weighted calculation on the abnormality degree, trend score and consistency index to obtain a loosening trend score; and generating an alarm information in the case that the loosening trend score is greater than or equal to a first threshold value. The method solves the problem that in the prior art, due to the complex underground environment, the change of wind resistance or the change of surrounding rock stress are easily disturbed, resulting in insufficient accuracy of early warning.

Description

Technical Field

[0001] This invention relates to the field of coal mine production safety technology, and more specifically, to a diagnostic method for loosening of mine structures. Background Technology

[0002] During mine mining, the stability of the surrounding rock and support structure directly affects ventilation safety and personnel safety. When the surrounding rock loosens, cracks expand, or the support deteriorates, it alters the effective ventilation cross-section and wall roughness, leading to a redistribution of local wind speed and stress fields, resulting in disturbances such as backflow, eddies, or abnormal wind speeds. These early structural changes are usually accompanied by weak aerodynamic noise and structural micro-vibrations, but their signal amplitude is small and their duration is short, making them easily masked by the complex noise environment underground (such as fan operation, equipment vibration, and personnel operations).

[0003] The main monitoring methods in the existing technology are, on the one hand, monitoring and simulating macroscopic parameters such as air volume, air pressure, and pressure difference at the ventilation network level. Although these methods can identify main fan anomalies or changes in damper status, they are difficult to distinguish subtle changes in wind resistance caused by local structural loosening. On the other hand, there are surrounding rock deformation monitoring technologies, such as peripheral displacement observation, anchor cable stress gauges, delamination instruments, and three-dimensional laser scanning. Although these methods can reflect geometric deformation, they have a lag in response, are not sensitive to early aerodynamic disturbances caused by fluid-structure interaction mechanisms, and require a long observation period to accumulate effective data.

[0004] In summary, existing detection methods are easily affected by the downhole working environment, resulting in insufficient sensitivity in monitoring early signs of loosening, changes in ventilation conditions, and structural disturbances, making it difficult to guarantee the accuracy of early warnings. Summary of the Invention

[0005] The main objective of this application is to provide a diagnostic method for loosening of mine structures, so as to at least solve the problem that the accuracy of early warning is insufficient due to the complexity of the underground environment and the susceptibility of interference based on changes in wind resistance or changes in surrounding rock stress.

[0006] To achieve the above objectives, according to one aspect of this application, a method for diagnosing mine structural loosening is provided, comprising: acquiring multi-source monitoring data and ventilation record data within a ventilation flow field, wherein the multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data; preprocessing the multi-source monitoring data and fusing the multi-source monitoring data and ventilation record data into fused monitoring data, wherein the preprocessing includes feature extraction, spatiotemporal alignment, and noise reduction; calculating the anomaly degree, trend score, and consistency index of the fused monitoring data, wherein the anomaly degree is determined using statistical methods or machine learning methods, and the trend score is determined using a time-series processing model, wherein the time-series processing model includes at least a long short-term memory model; performing a weighted calculation on the anomaly degree, trend score, and consistency index to obtain a loosening trend score; and generating an alarm message when the loosening trend score is greater than or equal to a first threshold.

[0007] Optionally, acquiring multi-source monitoring data and ventilation record data within the ventilation flow field includes: receiving monitoring signals from a first wind speed sensor group and a second wind speed sensor group to obtain wind speed monitoring data; the first wind speed sensor group is deployed at the monitoring section, and the second wind speed sensor group is deployed along the roadway axis; acquiring target acoustic signals through a sound pickup device to obtain sound disturbance data; the target acoustic signals are broadband acoustic signals generated by ventilation airflow passing through structural gaps, abrupt cross sections, and rough surfaces; the sound pickup device is deployed within a first preset range of the monitoring section; acquiring vibration signals and temperature signals through distributed optical fibers to obtain fiber optic sensing data; the distributed optical fibers are continuously deployed along one or more surfaces of the surrounding rock, support structure, anchor cables, and steel strips; and acquiring fan start / stop signals, damper opening / closing status, tunneling machine and transportation equipment operating status, and personnel work periods through a data interface to obtain ventilation record data.

[0008] Optionally, the multi-source monitoring data is preprocessed, including: bandpass filtering the wind speed monitoring data based on a first bandwidth, and smoothing the wind speed monitoring data using the moving average method; calculating the mean, variance, and turbulence intensity based on the wind speed monitoring data, and extracting the spectral features of the wind speed monitoring data using Fourier transform to obtain wind speed signal features; bandpass filtering the sound disturbance data based on a second bandwidth, determining the sound source of the sound disturbance data using array beamforming or time-difference localization, and denoising the sound disturbance data based on the sound source; processing the sound disturbance data using short-time Fourier transform, and extracting the Mel-frequency cepstral coefficients, spectral centroid, bandwidth, spectral kurtosis, and amplitude gradient of the sound disturbance data to obtain acoustic signal features; sequentially performing sliding window smoothing, temperature compensation, and baseline drift correction processing on the fiber optic sensing data; decomposing the fiber optic sensing data using wavelet transform, and extracting the energy distribution and impulse events of the fiber optic sensing data to obtain fiber optic disturbance features; and performing spatiotemporal alignment of the wind speed signal features, acoustic signal features, and fiber optic disturbance features.

[0009] Optionally, the multi-source monitoring data and ventilation record data are fused into fused monitoring data, including: using a gradient boosting tree model to determine the weights of each multi-source monitoring data and the ventilation record data to obtain a first target weight; performing weighted fusion of each multi-source monitoring data and ventilation record data based on the first target weight to obtain fused monitoring data; or using a deep multimodal fusion network to splice and fuse each multi-source monitoring data and ventilation record data to obtain fused monitoring data.

[0010] Optionally, after preprocessing the multi-source monitoring data, the method further includes: in the case of missing multi-source monitoring data, inputting the non-missing multi-source monitoring data and ventilation record data into the corresponding sub-discriminators to calculate the corresponding trend scores, thereby obtaining the trend sub-scores corresponding to each multi-source monitoring data and ventilation record data, wherein the sub-discriminators are obtained by training the time-series processing model on the dataset of the corresponding category; in the case of missing ventilation record data, inputting each multi-source monitoring data into the corresponding sub-discriminators to calculate the corresponding trend scores, thereby obtaining the corresponding trend sub-scores; and processing the trend sub-scores using weighted average, voting, or Bayesian fusion to obtain the trend scores of the fused monitoring data.

[0011] Optionally, the anomaly degree of the fused monitoring data can be calculated by: processing the multi-source monitoring data using Hotelling's T2 control chart to obtain the anomaly degree; or processing the multi-source monitoring data using the isolated forest algorithm to obtain the anomaly degree; or processing the multi-source monitoring data using a support vector machine algorithm to obtain the anomaly degree.

[0012] Optionally, the trend score of the fusion monitoring data is calculated by: using a long short-term memory network to model the trend of the fusion monitoring data within a first preset time period in the future to obtain a predicted trend sequence; or using a gated recurrent unit to model the trend of the fusion monitoring data within a first preset time period in the future to obtain a predicted trend sequence; and determining the trend score corresponding to the fusion monitoring data based on the predicted trend sequence.

[0013] Optionally, the consistency index of the fused monitoring data is calculated, including: calculating the Pearson correlation coefficient of the fused monitoring data to obtain the consistency index; or calculating the dynamic time regularization similarity of the fused monitoring data to obtain the consistency index; or calculating the window sign consistency rate of the fused monitoring data to obtain the consistency index.

[0014] Optionally, after acquiring ventilation record data, the method further includes: if the ventilation record data does not change for a second preset duration, constructing a data baseline model based on multi-source monitoring data within a third preset duration prior to the current time, wherein the third preset duration is less than the second preset duration; constructing a data baseline model based on the multi-source monitoring data, wherein the data baseline model includes the mean, standard deviation, and quantiles of each multi-source monitoring data; and updating the parameters in the baseline model according to a preset time decay coefficient if the ventilation record data changes.

[0015] Optionally, after acquiring multi-source monitoring data and ventilation record data within the ventilation flow field, the method further includes: acquiring radar point cloud data and flow field trajectory data when the confidence level of the wind speed monitoring data is less than a preset value; the flow field trajectory data is obtained by monitoring the particle trajectory of the tracer under the action of airflow; calculating the instantaneous velocity vector field of different regions of the cross-section based on the radar point cloud data and flow field trajectory data; determining the vortex area, backflow intensity, and velocity gradient peak value based on the instantaneous velocity vector field; determining the vortex disturbance index and velocity extraction anomaly index based on the vortex area, backflow intensity, and velocity gradient peak value; preprocessing the vortex disturbance index, velocity extraction anomaly index, multi-source monitoring data, and ventilation record data; and fusing the vortex disturbance index, velocity extraction anomaly index, multi-source monitoring data, and ventilation record data into fused monitoring data.

[0016] Applying the technical solution of this application, in the above-mentioned diagnostic method for loosening of mine structures, firstly, multi-source monitoring data and ventilation record data within the ventilation flow field are acquired. The multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data. Then, the multi-source monitoring data is preprocessed, and the multi-source monitoring data and ventilation record data are fused into fused monitoring data. The preprocessing includes feature extraction, spatiotemporal alignment, and noise reduction. Afterward, the anomaly degree, trend score, and consistency index of the fused monitoring data are calculated. The anomaly degree is determined using statistical or machine learning methods, and the trend score is determined using a time-series processing model, which includes at least a long short-term memory model. Then, the anomaly degree, trend score, and consistency index are weighted and calculated to obtain a loosening trend score. Finally, if the loosening trend score is greater than or equal to a first threshold, an alarm message is generated. This application ensures the comprehensiveness of the basis for loosening prediction by fusing multi-source data, and comprehensively analyzes the possibility of mine structure loosening (loosening trend score) from multiple perspectives, including spatial (anomaly degree), temporal (trend score) and data integrity (consistency), so as to provide early warning of the risk of mine structure loosening and improve the accuracy of early warning. Attached Figure Description

[0017] Figure 1 A hardware structure block diagram of a mobile terminal for a method of diagnosing loose mine structures provided in an embodiment of this application is shown.

[0018] Figure 2 A schematic flowchart of a method for diagnosing loosening of mine structures according to an embodiment of this application is shown.

[0019] Figure 3 A structural block diagram of a diagnostic device for loose mine structures provided according to an embodiment of this application is shown.

[0020] The above figures include the following reference numerals:

[0021] 102. Processor; 104. Memory; 106. Transmission device; 108. Input / output device. Detailed Implementation

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] As described in the background section, existing detection methods are easily affected by the underground working environment, resulting in insufficient sensitivity in monitoring early signs of loosening, changes in ventilation conditions, and structural disturbances, making it difficult to guarantee the accuracy of early warnings. To address the problem that existing technologies are easily affected by the complex underground environment, and that early warnings based on changes in wind resistance or surrounding rock stress are easily interfered with, leading to insufficient accuracy of early warnings, embodiments of this application provide a diagnostic method for loosening of mine structures.

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0027] The methods and embodiments provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for a method of diagnosing loose mine structures according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0028] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the mine structure loosening diagnosis method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the aforementioned networks may include wireless networks provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0029] This embodiment provides a diagnostic method for loosening of mine structures that runs on a mobile terminal, computer terminal, or similar computing device. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0030] Figure 2 This is a flowchart of a method for diagnosing loosening of mine structures according to an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0031] Step S201: Acquire multi-source monitoring data and ventilation record data within the ventilation flow field. The multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data.

[0032] Specifically, multi-source data reflecting ventilation flow field disturbances and their structural responses are collected, and multi-source raw data are output. The multi-source monitoring data includes wind speed monitoring data, acoustic disturbance data, and fiber optic sensing data. Ventilation record data is used to reflect the operating status of the ventilation system. By simultaneously collecting wind speed monitoring data, the temporal variation information of wind speed in the ventilation flow field is obtained; by collecting acoustic disturbance data, the acoustic signals generated by the interaction between airflow and structural surfaces are captured; and by collecting fiber optic sensing data, the micro-vibration and temperature responses along the fiber optic cable routing path are obtained.

[0033] Step S202: Preprocess the multi-source monitoring data and merge the multi-source monitoring data and ventilation record data into fused monitoring data. The preprocessing includes feature extraction, spatiotemporal alignment and noise reduction.

[0034] Specifically, based on a unified time / space reference, the original data from multiple sources are synchronized, denoised, and aligned to generate feature vectors that describe flow field disturbances and structural responses. Then, the feature vectors are fused from multiple sources to obtain fused monitoring data.

[0035] Step S203: Calculate the anomaly degree, trend score and consistency index of the fused monitoring data. The anomaly degree is determined by statistical methods or machine learning methods, and the trend score is determined by a time series processing model, which includes at least a long short-term memory model.

[0036] Specifically, anomaly, trend score, and consistency index are calculated by fusing monitoring data. Anomaly is a quantitative assessment of deviation features in multi-source monitoring data based on statistical or machine learning methods. Trend score is obtained by predicting the dynamic pattern of monitoring data evolution over time using a time-series processing model, which includes at least a long short-term memory model to capture and model long-term dependencies in the data. Consistency index is used to measure the consistency of responses from different monitoring sources to the same disturbance trend.

[0037] Step S204: Weight the anomaly score, trend score, and consistency index to obtain the loosening trend score;

[0038] Specifically, a comprehensive loosening trend score is formed by weighting anomaly degree, trend score, and consistency index according to preset weights. Among them, anomaly degree reflects the degree of deviation of multi-source sensor data from the normal baseline, trend score characterizes the dynamic evolution direction of the loosening development of surrounding rock or support structure, and consistency index measures the degree of synergistic support among three types of features: wind speed disturbance, acoustic disturbance, and fiber optic response for the same loosening trend.

[0039] Step S205: If the loosening trend score is greater than or equal to the first threshold, generate an alarm message.

[0040] Specifically, early warning levels are determined based on trend indices, and spatial locations of suspected areas are provided. Furthermore, when operating conditions change, the baselines for anomaly calculation and trend score calculation can be updated based on the early warning information to form a closed-loop correction for monitoring and judgment.

[0041] It is understandable that in the above embodiments, the disturbance-response relationship of the ventilation flow field is the core: slight loosening of the surrounding rock or support in the tunnel will change the effective ventilation cross-section, gap morphology and roughness, thereby causing measurable changes in the velocity field, pressure field and vortex structure; these changes will leave "fingerprints" in aerodynamic noise and structural micro-vibrations. The loosening trend is identified through the link of sensor data system → spatial synchronization and feature generation → fusion discrimination and trend prediction → early warning and location, and the early warning result triggers the baseline update to form a closed loop.

[0042] Specifically, local loosening leads to changes in the effective cross-section and an increase in roughness, resulting in an increase in equivalent ventilation resistance. Under near-steady-state conditions, the conservation of volumetric flow rate causes a redistribution of the average velocity and local velocity gradient, leading to changes in the amplitude and spectral structure of velocity fluctuations.

[0043] When gaps, steps, or abrupt cross sections appear, separation and reattachment cause measurable changes in the vortex area, backflow intensity, and peak velocity gradient; this effect can be used as a source of "vortex disturbance index / velocity gradient anomaly index" by cross-sectional field measurement or slice inversion.

[0044] Airflow scouring gaps and rough surfaces can generate broadband aerodynamic noise; when the geometric scale, incoming flow velocity and gap morphology change, the characteristic frequency band energy, spectral peaks and kurtosis in the sound spectrum also change accordingly, which can be used as early signs.

[0045] Airflow disturbances superimposed on equipment vibrations and environmental stresses will manifest as changes in micro-vibration energy and temperature gradients along the line in the surrounding rock / support-fiber coupling system; the fiber optic channel provides response curves on continuous spatial coordinates, which facilitates the location of the trend source.

[0046] All of the above abnormal changes can serve as data support for early warning. This application comprehensively considers the possibility of loosening the warning due to the above changes, and ensures the accuracy of the early warning.

[0047] In this embodiment, firstly, multi-source monitoring data and ventilation record data within the ventilation flow field are acquired. The multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data. Then, the multi-source monitoring data is preprocessed, and the multi-source monitoring data and ventilation record data are fused into fused monitoring data. Preprocessing includes feature extraction, spatiotemporal alignment, and noise reduction. Next, the anomaly degree, trend score, and consistency index of the fused monitoring data are calculated. The anomaly degree is determined using statistical or machine learning methods, and the trend score is determined using a time-series processing model, which includes at least a long short-term memory model. Then, the anomaly degree, trend score, and consistency index are weighted and calculated to obtain a loosening trend score. Finally, if the loosening trend score is greater than or equal to a first threshold, an alarm message is generated. This application ensures the comprehensiveness of the loosening prediction basis through multi-source data fusion and comprehensively analyzes the possibility of mine structure loosening (loosening trend score) from multiple perspectives, including space (anomaly degree), time (trend score), and data integrity (consistency), to provide early warning of mine structure loosening risks, thus improving the accuracy of the early warning.

[0048] In order to obtain the aforementioned multi-source monitoring data and ventilation record data, in an optional implementation, step S201 includes:

[0049] Step S2011: Receive monitoring signals from the first wind speed sensor group and the second wind speed sensor group to obtain wind speed monitoring data. The first wind speed sensor group is deployed at the monitoring section, and the second wind speed sensor group is deployed along the roadway axis.

[0050] Understandably, a first set of measuring points is set up on the monitoring section, and a second set of measuring points is set up along the roadway axis. The first set of measuring points (the aforementioned first wind speed sensor set) and the second set of measuring points (the aforementioned second wind speed sensor set) are synchronously sampled using a unified clock source to form wind speed time-series data covering both the cross-sectional distribution and the roadway distribution. Preferably, the first set of measuring points includes at least three measuring points, respectively arranged at the upper, lower, and side parts of the cross-section; the spacing between the measuring points in the second set of measuring points can be selected according to the roadway length and site conditions, preferably 1 to 5 meters. All measuring points are synchronously sampled using a unified clock source, and the sampling frequency is preferably not less than 10Hz to capture transient changes during ventilation fluctuations.

[0051] In practical implementation, wind speed measurement can be performed using an intrinsically safe ultrasonic anemometer for mining, which has the advantages of not needing to contact fluids, strong dust resistance, and high measurement accuracy. Alternatively, a pitot tube can be used in conjunction with an intrinsically safe differential pressure sensor for mining, and the wind speed can be obtained by measuring the dynamic pressure and combining it with the gas density. In the low-speed range or local vortex area, a hot film / hot wire anemometer with temperature compensation can also be selected for supplementary monitoring.

[0052] Step S2012: Acquire the target acoustic signal through the sound pickup device to obtain sound disturbance data. The target acoustic signal is a broadband acoustic signal generated by the ventilation airflow passing through structural gaps, abrupt cross sections and rough surfaces. The sound pickup device is deployed within the first preset range of the monitoring cross section.

[0053] Understandably, the sound pickup device is positioned near the monitoring section to acquire broadband acoustic signals generated by ventilation airflow passing through structural gaps, abrupt cross sections, and rough surfaces, thereby obtaining the aforementioned sound disturbance data.

[0054] In practical implementation, sound pickup devices can be arranged on both sides of the monitoring section, and one or two additional pickup points can be added at key locations along the downstream direction to form a small array structure. This can improve spatial coverage and provide conditions for directional sound source localization. The sound pickup unit can be a mining intrinsically safe MEMS microphone. This type of sensor has a compact structure, strong dust resistance, and sensitive response to broadband signals in the range of 20Hz to 5kHz. Alternatively, a mining intrinsically safe condenser or electret microphone can be selected according to the noise environment at the site, and equipped with a dustproof mesh cover or a waterproof acoustic membrane to reduce the impact of environmental factors on measurement accuracy.

[0055] Step S2013: Vibration and temperature signals are acquired through distributed optical fibers to obtain optical fiber sensing data. The distributed optical fibers are continuously deployed along one or more surfaces of the surrounding rock, support structure, anchor cable, and steel strip.

[0056] It is understandable that distributed optical fibers are continuously deployed along the surface of one of the surrounding rock, support components, anchor cables, or steel strips to acquire micro-vibration and temperature signals along the route. Based on the principle of backscattering distributed measurement, the micro-vibration and temperature on the optical fiber are demodulated to form a sequence of micro-vibration amplitude along the route. The sequence is then mapped to the response data along the route in spatial coordinates based on the optical fiber length calibration, thus obtaining the aforementioned optical fiber sensing data.

[0057] In practical implementation, single-mode armored optical fiber is preferred for distributed optical fiber deployment. The sheath material can be low-smoke, halogen-free, and flame-retardant to meet the requirements of mine fire prevention and mechanical protection. The fiber optic laying method can be flexibly selected according to the monitoring object: when monitoring surrounding rock, it can be laid along the roadway wall and fixed with expansion bolts or special clips; when monitoring support components, anchor cables, or steel strips, it can be laid along the outline of the component and secured with cable ties or metal clamps to ensure good mechanical coupling between the optical fiber and the monitored structure. Anchor points should be set at intervals of 10–30 meters, and folded-back wiring can be used at key nodes to enhance positioning accuracy.

[0058] The demodulation unit can employ distributed acoustic sensing (DAS, based on phase-sensitive optical time-domain reflectometry (Φ-OTDR)) to acquire the amplitude sequence of micro-vibrations along the line, with a spatial resolution of 1–10 meters and a temporal resolution preferably of 1–5 seconds. Alternatively, it can use distributed temperature sensing (DTS, based on Raman scattering time-domain reflectometry) to acquire the temperature sequence along the line, which can be used to analyze the rate of temperature rise and temperature gradient changes, and provide a basis for wind speed conversion and temperature drift compensation. If necessary, Brillouin scattering type (BOTDR / BOTDA) equipment can also be used to acquire the strain distribution along the line, enabling long-term structural deformation monitoring.

[0059] Based on the calibration relationship between fiber length and actual spatial coordinates, the positioning and aggregation unit maps the demodulated micro-vibration amplitude sequence and temperature sequence into line-side response data in spatial coordinates.

[0060] Step S2014: Obtain ventilation record data by acquiring fan start / stop signals, damper opening / closing status, tunneling machine and transportation equipment operating status, and personnel working hours through the data interface.

[0061] Understandably, recording operating condition labels such as fan start / stop, damper opening / closing, ventilation scheduling instructions, mining equipment operating status, and personnel working periods, and aligning them with wind speed time series data, can be used as baseline modeling and feature correction in subsequent early warning processes.

[0062] In practice, the system can automatically acquire information such as the start and stop signals of the main fan and local fans, the opening and closing status of the ventilation doors, and the operating status of the tunneling machine and transportation equipment through the data interface with the mine monitoring system. For information that can not be manually intervened or automatically collected, such as personnel working time, it can be entered and uploaded to the system through a handheld terminal.

[0063] In the above embodiments, a mine time server can be selected as the unified clock source to distribute time signals to each sensor node via fiber optic or wired networks. For applications requiring higher precision synchronization, IEEE 1588 Precision Time Protocol (PTP) or IRIG-B code time synchronization technology can be used, supplemented by GPS time synchronization (if mine conditions permit) for time synchronization on the ground. The time is then transmitted underground to each acquisition unit via fiber optic network, thereby achieving millisecond-level or even higher precision system-wide synchronization. The spatial reference unit establishes a coordinate system based on the geometric layout of the mine roadway, using the mileage markers on the roadway centerline as the longitudinal reference and the cross-sectional center point as the lateral reference. Combined with fiber optic cable laying length calibration, wind speed measurement point location recording, and acoustic pickup unit layout diagram, the unique positioning of each sensor in spatial coordinates is achieved.

[0064] In order to perform feature extraction and spatiotemporal alignment on the above-mentioned multi-source monitoring data, in an optional implementation, step S202 includes:

[0065] Step S20201: Bandpass filtering is performed on the wind speed monitoring data based on the first bandwidth, and the moving average method is used to smooth the wind speed monitoring data.

[0066] Understandably, calling the digital signal processing (DSP) module first removes high-frequency noise through low-pass filtering, and then uses a Kalman filter to smooth fluctuations. The principle behind this is to perform recursive optimal estimation of the measured values ​​based on a state-space model.

[0067] Step S20202: Calculate the mean, variance, and turbulence intensity based on the wind speed monitoring data, and use Fourier transform to extract the spectral features of the wind speed monitoring data to obtain the wind speed signal features;

[0068] Step S20203: Bandpass filtering is performed on the sound disturbance data based on the second bandwidth, and the sound source of the sound disturbance data is determined by array beamforming or time difference localization, and the sound disturbance data is denoised based on the sound source.

[0069] Understandably, the acquired broadband acoustic signals are processed by a processing unit to perform bandpass filtering (e.g., 100Hz to 3000Hz) under a unified time reference to suppress low-frequency mechanical vibrations and high-frequency electromagnetic interference. Furthermore, the direction of the main sound source can be estimated based on array beamforming or time difference positioning to obtain the sound source direction estimation result.

[0070] In practice, array beamforming technology or time delay difference-based positioning algorithms are used to estimate the direction of the main sound source to help determine the spatial correlation between noise changes and potential structural loosening.

[0071] Step S20204: The sound disturbance data is processed by short-time Fourier transform, and the Mel-frequency cepstral coefficients, spectral centroid, bandwidth, spectral kurtosis and amplitude gradient of the sound disturbance data are extracted to obtain the acoustic signal characteristics.

[0072] Furthermore, under a unified time reference, bandpass filtering, short-time Fourier transform, or wavelet packet decomposition are performed on the broadband acoustic signal to decompose the signal into a feature representation in both the frequency and time domains. By performing statistical calculations on the spectral data (including spectral subtraction to reduce background noise interference), acoustic perturbation features such as spectral peak frequency, in-band energy ratio, spectral kurtosis, spectral centroid, and characteristic band amplitude gradient are extracted to generate acoustic feature vectors.

[0073] Step S20205: Perform sliding window smoothing, temperature compensation, and baseline drift correction on the fiber optic sensing data in sequence.

[0074] Specifically, using the principle of phase-sensitive distributed optical fiber sensing (Φ-OTDR), optical pulses are injected into the optical fiber, and the location and intensity of micro-vibrations are obtained through Rayleigh scattering echo analysis; the acquired raw waveforms are transformed by Hilbert to obtain the envelope feature sequence.

[0075] Step S20206: Wavelet transform is used to decompose the fiber optic sensing data, and the energy distribution and pulse events of the fiber optic sensing data are extracted to obtain the fiber optic disturbance characteristics.

[0076] Step S20207: Spatiotemporal alignment of wind speed signal characteristics, acoustic signal characteristics, and fiber optic disturbance characteristics.

[0077] In practical implementation, since different sensors may have different sampling rates, the system uses linear interpolation or spline interpolation methods to resample the data to a uniform sampling frequency to achieve precise alignment. Then, the various types of data after noise reduction are matched according to a uniform timestamp, and the data are mapped to the same spatial reference frame according to the spatial coordinate system to ensure the spatiotemporal correspondence of multi-source data.

[0078] In the above embodiments, various feature vectors can be standardized (Z-score Normalization), and principal component analysis (PCA) can be used to reduce redundant dimensions, which is equivalent to feature engineering. That is, to achieve unified encoding of multi-channel sensor data in time and space, so that the data of different physical quantities are comparable and fusionable, providing high-quality input for subsequent fusion discrimination and trend prediction.

[0079] To integrate the aforementioned multi-source monitoring data, in an optional implementation, step S202 further includes:

[0080] Step S20209: Use the gradient boosting tree model to determine the weights of each multi-source monitoring data and the weights of the ventilation record data to obtain the first target weights;

[0081] Step S20210: Based on the first objective weight, the multi-source monitoring data and ventilation record data are weighted and fused to obtain fused monitoring data; or

[0082] Step S20211: A deep multimodal fusion network is used to stitch together and fuse the multi-source monitoring data and ventilation record data to obtain fused monitoring data.

[0083] In the above embodiments, feature weight coefficients are set according to the importance and historical performance of each channel, and weighted concatenation is performed to generate a comprehensive feature vector. The weight coefficients can be dynamically adjusted through offline training or online learning, for example, by using a weighted regression algorithm based on historical prediction accuracy. Specifically, a weighted feature fusion method is used to linearly combine sub-features such as wind speed, acoustics, and fiber optic disturbance according to their importance weights. The weights are determined by an offline-trained gradient boosting tree (GBDT) model. Alternatively, a deep multimodal fusion network is used, in which convolutional or recurrent layers of different modalities are set in the neural network, and then feature concatenation and fusion are performed in a fully connected layer.

[0084] To integrate the aforementioned multi-source monitoring data, in one optional implementation, after preprocessing the multi-source monitoring data, the method further includes:

[0085] Step S301: In the case of missing multi-source monitoring data, input the non-missing multi-source monitoring data and ventilation record data into the corresponding sub-discriminators to calculate the corresponding trend scores, and obtain the trend sub-scores corresponding to each multi-source monitoring data and ventilation record data. The sub-discriminators are obtained by training the time series processing model on the dataset of the corresponding category.

[0086] Step S302: In the case of missing ventilation record data, input the multi-source monitoring data into the corresponding sub-discriminator to calculate the corresponding trend score and obtain the corresponding trend sub-score.

[0087] Step S303: The trend sub-scores are processed by weighted average, voting, or Bayesian fusion to obtain the trend scores of the fused monitoring data.

[0088] In the above embodiments, each sub-vector can be individually input into its corresponding sub-discriminator (such as a model based on support vector machine, random forest, convolutional neural network, etc.) to obtain its own trend index prediction result. Then, the outputs of these sub-discriminators are weighted and averaged, voted on, or fused based on Bayesian updates to obtain the final trend index. When some channel data is missing, decision-level fusion can automatically ignore the impact of missing channels while ensuring the overall prediction.

[0089] In order to analyze the potential loosening risk from a spatial perspective, in one optional implementation, step S203 above includes:

[0090] Step S2031: Process the multi-source monitoring data using Hotelling's T2 control chart to obtain the anomaly degree; or

[0091] Step S2032: The Isolation Forest algorithm is used to process the multi-source monitoring data to obtain the anomaly score; or

[0092] Step S2033: A support vector machine algorithm is used to process the multi-source monitoring data to obtain the anomaly degree.

[0093] In the above embodiments, statistical methods are used: based on Hotelling's T² control chart, the multidimensional feature vector is detected to determine whether it exceeds the normal range in order to determine the above-mentioned anomaly degree; or an isolation forest or one-class support vector machine is used to identify rare patterns to obtain the above-mentioned anomaly degree.

[0094] In order to analyze the potential risk of loosening from a time perspective, in an optional implementation, step S203 above includes:

[0095] Step S2033: Use a Long Short-Term Memory (LSTM) network to model the changing trend of the fused monitoring data within a first preset time period in the future, and obtain a predicted trend sequence; or

[0096] Step S2034: A gated loop unit is used to model the changing trend of the fused monitoring data within a first preset time period in the future to obtain a predicted trend sequence;

[0097] Step S2035: Determine the trend score corresponding to the fused monitoring data based on the predicted trend sequence.

[0098] In the above embodiments, time series analysis models (such as Long Short-Term Memory Network LSTM, Temporal Convolutional Network TCN, Kalman Filter predictor, etc.) are used to predict the loosening trend in the future and output the trend index and the corresponding confidence parameters.

[0099] Understandably, the model establishes a baseline based on steady-state data to ensure that the predicted reference values ​​are consistent with the actual mine operating environment.

[0100] In order to analyze the potential risk of loosening from a time perspective, in an optional implementation, step S203 above includes:

[0101] Calculate the Pearson correlation coefficient of the fused monitoring data to obtain the consistency index; or

[0102] Calculate the dynamic time-warped similarity of the fused monitoring data to obtain a consistency index;

[0103] The consistency rate of window symbols in the fused monitoring data is calculated to obtain the consistency index.

[0104] In the above embodiments, statistical consistency indices among multi-source monitoring data are obtained by calculating the Pearson correlation coefficient, dynamic time regularization similarity, or window sign consistency rate of the fused monitoring data. These indices are then weighted and comprehensively evaluated in conjunction with anomaly and trend scores. This effectively distinguishes between non-structural ventilation disturbances caused by single-channel interference and genuine structural loosening signs resulting from multi-source coordinated responses. When wind speed, sound disturbances, and fiber optic sensing data in the ventilation flow field exhibit significant coordinated changes in temporal characteristics, distribution patterns, or sign patterns, the consistency index will significantly increase, thereby enhancing the confidence level of the loosening trend score. Conversely, if only one data point shows anomalies while other channels do not respond synchronously, the consistency index decreases, suppressing false alarms and enabling accurate identification and reliable diagnosis of early loosening of mine structures under strong interference environments.

[0105] In practice, a consistency index is calculated among the prediction results of different channels to reflect the degree to which multi-source information supports the same trend in the current time period. The consistency index can be calculated using methods such as Pearson correlation coefficient, dynamic time warping (DTW) similarity, or window-based sign consistency rate. When the consistency index is higher than a preset threshold (e.g., 0.7–0.8), it indicates that the judgments of each channel on the trend are relatively consistent, which can improve the confidence of the final result. When the consistency index is lower than the threshold, the system will automatically reduce the weight of single-source anomalies to prevent noise or outliers in individual channels from affecting the overall judgment.

[0106] To ensure the accuracy of the early warning, in one optional implementation, after acquiring the ventilation record data, the above method further includes:

[0107] Step S401: If the ventilation record data does not change for a second preset duration, based on the multi-source monitoring data within a third preset duration before the current moment, the third preset duration is less than the second preset duration;

[0108] Step S402: Construct a data baseline model based on multi-source monitoring data. The data baseline model includes the mean, standard deviation, and quantiles of each multi-source monitoring data.

[0109] Step S403: If the ventilation record data changes, update the parameters in the baseline model according to the preset time decay coefficient.

[0110] In the above embodiments, during periods of stable mine ventilation and production conditions, the system collects multi-source feature vectors such as wind speed disturbances, acoustic disturbances, and fiber optic responses over a period of time (e.g., 24 hours or longer), and calculates statistical quantities such as the mean, standard deviation, and quantiles of each feature. These statistical results are stored as a baseline model, serving as a reference standard for subsequent trend index calculation and threshold judgment. Once events such as main fan start-up / stop, damper status change, mining equipment start-up / stop, or personnel operation are detected, it is determined that the current operating condition has changed. After the operating condition changes, the system gradually updates the baseline parameters according to a preset time decay coefficient (e.g., exponential decay coefficient α = 0.8 to 0.95), so that the new operating state is included in the baseline range within a certain period of time, thereby avoiding false alarms caused by short-term operating condition fluctuations.

[0111] Furthermore, the adaptive adjustment of the threshold can be dynamically adjusted based on the feature distribution calculated by the sliding window. For example, the warning threshold can be set as the baseline mean plus K times the standard deviation, and the value of K can be optimized according to the importance and historical performance of different channels.

[0112] Through the above embodiments, the system establishes a baseline model during the steady-state period to identify changes in operating conditions such as fan start-up and shutdown, damper opening and closing, mining equipment operation, and personnel work periods; after changes in operating conditions, the baseline and threshold are updated according to a preset time decay coefficient; when the early warning and positioning subsystems generate continuous early warnings in the same spatial area, local baseline re-estimation and sensor self-check prompts are triggered to reduce the impact of unstructured disturbances and correct subsequent identification results.

[0113] The system establishes a baseline model during steady-state operation and adaptively adjusts the threshold according to changes in operating conditions during operation. When a persistent warning occurs, it also triggers local baseline reassessment and sensor self-check, thus forming a closed-loop mechanism of identification-warning-correction to improve the reliability of trend identification.

[0114] The baseline model also records operating condition descriptions to distinguish benchmark values ​​under different operating conditions. For example, different main fan speeds and different damper opening and closing states can each have corresponding baselines established to ensure consistent environmental conditions during comparison.

[0115] To ensure the accuracy of the early warning, in one optional implementation, after acquiring multi-source monitoring data and ventilation record data within the ventilation flow field, the above method further includes:

[0116] Step S501: If the confidence level of the wind speed monitoring data is less than a preset value, acquire radar point cloud data and flow field trajectory data. The flow field trajectory data is obtained by monitoring the particle trajectory of the tracer under the action of air flow.

[0117] Understandably, the point cloud and tracer trajectory of the monitoring section are obtained, and the point cloud and trajectory are registered and inverted to form vortex disturbance index and velocity gradient anomaly index.

[0118] In practical implementation, intrinsically safe two-dimensional or multi-line laser scanning equipment for mining can be selected. Its measurement range should cover the full size of the monitored cross-section, with an angular resolution preferably of 0.1° to 0.5° and a scanning frequency preferably of 5 to 15 Hz, to ensure that the complete cross-sectional geometry and tracer particle distribution can be captured within the ventilation flow field fluctuation cycle. To adapt to the dusty and humid environment underground, the lidar housing should have a dustproof and waterproof rating (such as IP65 or above), and can be equipped with an air curtain blowing device or a dustproof window to keep the optical window clean.

[0119] The tracer delivery module can select the tracer medium according to safety and visibility requirements, such as food-grade glycerin atomized liquid, fine water mist, or other approved inert aerosols. The tracer medium is uniformly released into the upstream airflow through the nozzle, and the release time can be controlled within 2 to 5 seconds to form a recognizable particle swarm for flow field tracking.

[0120] Step S502: Calculate the instantaneous velocity vector field of different regions of the cross section based on radar point cloud data and flow field trajectory data, and determine the vortex area, backflow intensity and peak velocity gradient based on the instantaneous velocity vector field;

[0121] Step S503: Determine the vortex disturbance index and velocity extraction anomaly index based on the vortex area, backflow intensity, and peak velocity gradient.

[0122] Step S504: Preprocess the vortex disturbance index, velocity extraction anomaly index, multi-source monitoring data and ventilation record data, and merge the vortex disturbance index, velocity extraction anomaly index, multi-source monitoring data and ventilation record data into fused monitoring data.

[0123] In the above embodiments, point cloud data acquired by lidar is registered with tracer images (or particle trajectory data). Particle image velocimetry (PIV) or trajectory-based velocity inversion algorithms are used to calculate the instantaneous velocity vector field in different regions of the cross-section. Parameters such as vortex area, backflow intensity, and peak velocity gradient are extracted to form vortex disturbance indices and velocity gradient anomaly indices. When a wind speed measurement module is not configured, these indices can be directly input as cross-sectional flow field disturbance parameters into subsequent analysis processes. When a wind speed measurement module is configured, they are input in parallel with wind speed data to improve the spatial resolution and comprehensiveness of the cross-sectional disturbance parameters.

[0124] In one embodiment of this application, the system also adaptively updates based on the warning results. Specifically, when the system generates high-level warnings for multiple consecutive time periods (e.g., more than 5 minutes) within the same spatial area, the system marks that area as a key review area and triggers local baseline reassessment. During the local baseline reassessment process, the system recalculates the baseline parameters of the area by combining historical data and current stable period data.

[0125] At the same time, the system will also send sensor self-test requests to the maintenance terminal to perform functional tests on channels such as wind speed, acoustics, and fiber optics in key areas to prevent false warnings caused by sensor failures.

[0126] This closed-loop mechanism ensures that the system can quickly adjust its judgment criteria and reduce false alarm rates when faced with non-structural disturbances (such as short-term ventilation adjustments or construction operations); at the same time, it can quickly lock and accurately locate the problem when a real structural loosening trend appears by continuously issuing warnings and comparing with the baseline.

[0127] In another embodiment of this application, the system can also locate the loosening risk based on the early warning information, wherein the location of the event source is calculated based on the difference in disturbance intensity at multiple measurement points and the time difference of arrival (TDOA), and the precise location is determined by combining the propagation time and reflection characteristics of the Φ-OTDR signal along the optical fiber.

[0128] To enable those skilled in the art to better understand the technical solution of this application, the implementation process of the method for diagnosing loose mine structures will be described in detail below with reference to specific embodiments.

[0129] In one specific embodiment, a multi-source fusion master scheme (wind speed array + acoustic + distributed optical fiber) is adopted. It is suitable for mining roadways or connecting roadways with stable ventilation, moderate dust, and regular roadway cross-sections, with the goal of identifying and locating the loosening trend of surrounding rock or support at an early stage.

[0130] Further configuration includes a wind speed measurement module (section group + friction group). The section group involves setting at least three measuring points (upper, lower, and side) on the target monitoring section; the distance from each measuring point to the roadway wall is 0.3–0.5m. The friction group involves placing measuring points along the roadway axis from upstream to downstream, with a spacing of 2–3m, covering 15–30m. The sensor is a mining-grade intrinsically safe ultrasonic anemometer (range 0–20m / s, resolution 0.01m / s, accuracy ≤±2%FS), with all measuring points sampled synchronously using a unified clock. Sampling frequency is 10–20Hz; each analysis window is 60s, with an overlap ratio of 50%. Pitot tubes and intrinsically safe differential pressure (0–500Pa) can be added to key sections for cross-checking the average wind speed of the section. A condition marking module is included, connecting to main / local fan start / stop signals, damper position switches, and tunneling equipment operating status (485 / Ethernet), manually entered by the work team during work periods. All data is uniformly written to a condition timestamp queue for baseline and threshold adaptive use. Aeroacoustic acquisition module (small array), layout: 1 point on each side of the monitoring section, and an additional point 5–m downstream, for a total of 3 small array points; equipped with a dustproof acoustic mesh cover. Sensor: Intrinsically safe MEMS microphone, frequency 20Hz–5kHz. Sampling: 8kHz, frame length 1024 points, frame shift 256 points. Distributed fiber optic sensing module, fiber: 9 / 125μm single-mode armored fiber, continuously laid for 50–120m along the right wall of the tunnel at a height of 0.8–1.2m, mechanically anchored every 15–20m, with a "Z" shaped fold in the key area. Demodulation: DAS ( -OTDR) acquires the micro-vibration intensity sequence along the line (spatial resolution 2–5m, time resolution 1–2s); DTS in parallel with the fiber optic channel acquires the temperature sequence (optional). Time synchronization: PTP time synchronization is used between the fiber optic demodulator and the wind speed and acoustic acquisition host, with drift ≤1ms.

[0131] Further processing, synchronization, and feature generation: Time alignment: The raw data from each channel are first aligned to a unified time axis; different sampling rates are interpolated using splines to a unified time step (e.g., 10Hz). Preprocessing: Wind speed: 0.05–2Hz bandpass filter + median filtering + outlier removal. Acoustics: 100–3000Hz bandpass filter, STFT with Hamming window to obtain time spectrum. Fiber optics: Moving average (3–5 points) + temperature drift compensation. Feature construction (60s window, 30s step): Wind speed sub-vector: mean μ, variance σ², turbulence intensity TI = σ / μ, path gradient G, short-time energy E. Acoustic sub-vector: MFCC (13-dimensional), spectral kurtosis SK, spectral centroid SC, in-band energy ratio Rb. Fiber optic sub-vector: segment energy EL (L for 5–10m segments), temperature gradient dT / dx (if DTS is used). Unified Z-score normalization; PCA retains 95% variance. Fusion and Trend Prediction: Feature-level fusion: After concatenating the vectors, embedding is performed through a fully connected layer (128→64), ReLU activation, and Dropout 0.2. Anomaly detection: IsolationForest (n_estimators=200, max_samples="auto") outputs anomaly score a∈[0,1]. Trend prediction: Two-layer LSTM (64→32), input is the fusion vector of the most recent N=20 frames, output is the trend score S∈[0,1] for the next 5 minutes. Consistency index: DTW similarity between wind speed and acoustics, and wind speed and optical fiber is calculated within the same window, and the mean C∈[0,1] is used as consistency. Loosening trend index: I=σ(w_1S+w_2a+w_3c), w_1=0.5, w_2=0.3, w_3=0.2. Where σ is Sigmoid. The weights can be fine-tuned according to the historical playback of the site.

[0132] Further, early warning and location: Grading thresholds (from baseline statistics during the steady-state period of 7–14 days): Level I: I ≥ p_90, and C ≥ 0.7; Level II: I ≥ p_95, and C ≥ 0.7; Level III: I ≥ p_98, and C ≥ 0.75. Location: Along-line location: Take the center coordinates of the local peak section of the fiber optic EL; Cross-sectional location: Within the same time window, take the centroid of the measurement point group with the largest rising / falling edge of the cross-sectional wind speed field; Fusion location: Weight the along-line coordinates and the cross-sectional centroid with 0.6 / 0.4 to obtain the final suspected location. Linkage: Level II and above will display a heat map and issue a voice alarm on the monitoring terminal, pushing dispatch information.

[0133] Baseline and Closed Loop: Baseline Establishment: Select days without construction disturbance, record continuously for ≥7 days, and generate μ, σ, and quantiles for each feature. Adaptation: After detecting damper switching or changes in main fan operating conditions, update the threshold using EWMA with coefficient α=0.9. Continuous Early Warning Processing: If Level II or above is triggered for three consecutive windows in the same coordinate area, the system automatically initiates local baseline reassessment and sensor self-check prompts.

[0134] In another specific embodiment, a cross-sectional slicing alternative (LiDAR + tracing + inversion) is used. This is necessary for locations with significant cross-sectional geometric changes, bends, and confluences of connecting lanes, where visualization of cross-sectional flow patterns and identification of vortex / recirculation and velocity gradient anomalies are required. This is particularly relevant when wind speed array placement is limited, or when it is desirable to obtain high-resolution cross-sectional disturbance indicators with less hardware.

[0135] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0136] This application also provides a diagnostic device for mine structure loosening. It should be noted that this diagnostic device can be used to execute the diagnostic method for mine structure loosening provided in this application. This device is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0137] The following describes the diagnostic device for loose mine structures provided in the embodiments of this application.

[0138] Figure 3 This is a structural block diagram of a diagnostic device for loose mine structures according to an embodiment of this application. Figure 3 As shown, the device includes:

[0139] The first acquisition unit 10 is used to acquire multi-source monitoring data and ventilation record data in the ventilation flow field. The multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data.

[0140] The first processing unit 20 is used to preprocess the multi-source monitoring data and merge the multi-source monitoring data and ventilation record data into fused monitoring data. The preprocessing includes feature extraction, spatiotemporal alignment and noise reduction.

[0141] The first calculation unit 30 is used to calculate the anomaly degree, trend score and consistency index of the fused monitoring data. The anomaly degree is determined by statistical methods or machine learning methods, and the trend score is determined by a time series processing model. The time series processing model includes at least a long short-term memory model.

[0142] The second calculation unit 40 is used to perform weighted calculations on the abnormality, trend score and consistency index to obtain the loosening trend score;

[0143] The generation unit 50 is used to generate alarm information when the loosening trend score is greater than or equal to a first threshold.

[0144] In this embodiment, the first acquisition unit acquires multi-source monitoring data and ventilation record data within the ventilation flow field. The multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data. The first processing unit preprocesses the multi-source monitoring data and merges the multi-source monitoring data and ventilation record data into fused monitoring data. The preprocessing includes feature extraction, spatiotemporal alignment, and noise reduction. The first calculation unit calculates the anomaly degree, trend score, and consistency index of the fused monitoring data. The anomaly degree is determined using statistical or machine learning methods, and the trend score is determined using a time-series processing model, which includes at least a long short-term memory model. The second calculation unit performs a weighted calculation on the anomaly degree, trend score, and consistency index to obtain a loosening trend score. The generation unit generates an alarm message when the loosening trend score is greater than or equal to a first threshold. This application ensures the comprehensiveness of the loosening prediction basis through multi-source data fusion and comprehensively analyzes the possibility of mine structure loosening (loosening trend score) from multiple perspectives, including space (anomaly degree), time (trend score), and data integrity (consistency), to provide early warning of the risk of mine structure loosening and improve the accuracy of the early warning.

[0145] To acquire the aforementioned multi-source monitoring data and ventilation record data, in one optional embodiment, the first acquisition unit includes:

[0146] The first acquisition module is used to receive monitoring signals from the first wind speed sensor group and the second wind speed sensor group to obtain wind speed monitoring data. The first wind speed sensor group is deployed at the monitoring section, and the second wind speed sensor group is deployed along the roadway axis.

[0147] The second acquisition module is used to acquire the target acoustic signal through the sound pickup device to obtain sound disturbance data. The target acoustic signal is a broadband acoustic signal generated by the ventilation airflow passing through structural gaps, abrupt cross sections and rough surfaces. The sound pickup device is deployed within the first preset range of the monitoring cross section.

[0148] The third acquisition module is used to acquire vibration and temperature signals through distributed optical fibers to obtain optical fiber sensing data. The distributed optical fibers are continuously deployed along one or more surfaces of the surrounding rock, support structure, anchor cable and steel strip.

[0149] The fourth acquisition module is used to acquire fan start / stop signals, damper opening / closing status, tunneling machine and transportation equipment operating status, and personnel working hours through the data interface to obtain ventilation record data.

[0150] In order to perform feature extraction and spatiotemporal alignment on the aforementioned multi-source monitoring data, in an optional implementation, the first processing unit includes:

[0151] The first processing module is used to perform bandpass filtering on the wind speed monitoring data based on the first bandwidth, and to smooth the wind speed monitoring data using the moving average method.

[0152] The second processing module is used to calculate the mean, variance, and turbulence intensity based on wind speed monitoring data, and to extract the spectral features of the wind speed monitoring data using Fourier transform to obtain the wind speed signal features.

[0153] The third processing module is used to perform bandpass filtering on the sound disturbance data based on the second bandwidth, determine the sound source of the sound disturbance data using array beamforming or time difference localization, and denoise the sound disturbance data based on the sound source.

[0154] The fourth processing module is used to process the sound disturbance data using short-time Fourier transform and extract the Mel-frequency cepstral coefficients, spectral centroid, bandwidth, spectral kurtosis and amplitude gradient of the sound disturbance data to obtain acoustic signal characteristics.

[0155] The fifth processing module is used to sequentially perform sliding window smoothing, temperature compensation, and baseline drift correction on the fiber optic sensing data.

[0156] The sixth processing module is used to decompose the fiber optic sensing data using wavelet transform and extract the energy distribution and pulse events of the fiber optic sensing data to obtain the fiber optic perturbation characteristics.

[0157] The seventh processing module is used for spatiotemporal alignment of wind speed signal characteristics, acoustic signal characteristics, and fiber optic disturbance characteristics.

[0158] To integrate the aforementioned multi-source monitoring data, in one optional implementation, the first processing unit further includes:

[0159] The first determination module is used to determine the weights of each multi-source monitoring data and the weights of ventilation record data using a gradient boosting tree model, so as to obtain the first target weights.

[0160] The tenth processing module is used to weight and fuse the multi-source monitoring data and ventilation record data based on the first objective weight to obtain fused monitoring data; or

[0161] The eleventh processing module is used to stitch together and fuse the multi-source monitoring data and ventilation record data using a deep multimodal fusion network to obtain fused monitoring data.

[0162] To integrate the aforementioned multi-source monitoring data, in one optional embodiment, the device further includes:

[0163] The third calculation unit is used to preprocess the multi-source monitoring data and, in the case of missing multi-source monitoring data, input the non-missing multi-source monitoring data and ventilation record data into the corresponding sub-discriminators to calculate the corresponding trend scores, thereby obtaining the trend sub-scores corresponding to each multi-source monitoring data and ventilation record data. The sub-discriminators are obtained by training the time series processing model on the dataset of the corresponding category.

[0164] The fourth calculation unit is used to input the multi-source monitoring data into the corresponding sub-discriminator to calculate the corresponding trend score when ventilation record data is missing, and obtain the corresponding trend sub-score.

[0165] The second processing unit is used to process the trend sub-scores by weighted average, voting, or Bayesian fusion to obtain the trend scores of the fused monitoring data.

[0166] To analyze potential loosening risks from a spatial perspective, in one optional implementation, the first computing unit includes:

[0167] The first calculation module is used to process multi-source monitoring data using Hotelling's T2 control charts to obtain the anomaly degree; or

[0168] The second calculation module is used to process multi-source monitoring data using the isolated forest algorithm to obtain anomaly scores; or

[0169] The third calculation module is used to process multi-source monitoring data using a support vector machine algorithm to obtain the anomaly degree.

[0170] To analyze potential loosening risks from a time perspective, in one optional implementation, the first calculation unit includes:

[0171] The twelfth processing module is used to model the changing trend of the fused monitoring data within a first preset time period using a long short-term memory network, to obtain a predicted trend sequence; or

[0172] The thirteenth processing module is used to model the changing trend of the fused monitoring data within a first preset time period using a gated loop unit, and obtain a predicted trend sequence.

[0173] The second determination module is used to determine the trend score corresponding to the fused monitoring data based on the predicted trend sequence.

[0174] To analyze potential loosening risks from a time perspective, in one optional implementation, the first calculation unit includes:

[0175] The fourth calculation module is used to calculate the Pearson correlation coefficient of the fused monitoring data to obtain the consistency index; or

[0176] The fifth calculation module is used to calculate the dynamic time-warped similarity of the fused monitoring data to obtain the consistency index;

[0177] The sixth calculation module is used to calculate the window symbol consistency rate of the fused monitoring data to obtain the consistency index.

[0178] To ensure the accuracy of the early warning, in one optional embodiment, the above-mentioned device further includes:

[0179] The second acquisition unit is used to, after acquiring ventilation record data, and provided that the ventilation record data does not change for a second preset duration, to acquire multi-source monitoring data within a third preset duration prior to the current moment, where the third preset duration is less than the second preset duration;

[0180] The fifth calculation unit is used to construct a data baseline model based on multi-source monitoring data. The data baseline model includes the mean, standard deviation, and quantiles of each multi-source monitoring data.

[0181] The sixth calculation unit is used to update the parameters in the baseline model according to a preset time decay coefficient when the ventilation record data changes.

[0182] To ensure the accuracy of the early warning, in one optional implementation, the above method further includes:

[0183] The third acquisition unit is used to acquire radar point cloud data and flow field trajectory data after acquiring multi-source monitoring data and ventilation record data in the ventilation flow field, when the confidence level of the wind speed monitoring data is less than a preset value. The flow field trajectory data is obtained by monitoring the particle trajectory of the tracer under the action of air flow.

[0184] The seventh calculation unit is used to calculate the instantaneous velocity vector field of different regions of the cross section based on radar point cloud data and flow field trajectory data, and to determine the vortex area, backflow intensity and peak velocity gradient based on the instantaneous velocity vector field.

[0185] The determination unit is used to determine the vortex disturbance index and the velocity extraction anomaly index based on the vortex area, backflow intensity and velocity gradient peak value.

[0186] The third processing unit is used to preprocess the vortex disturbance index, velocity extraction anomaly index, multi-source monitoring data and ventilation record data, and to merge the vortex disturbance index, velocity extraction anomaly index, multi-source monitoring data and ventilation record data into fused monitoring data.

[0187] The aforementioned diagnostic device for loose mine structures includes a processor and a memory. The first acquisition unit, first processing unit, first calculation unit, second calculation unit, and generation unit are all stored as program units in the memory. The processor executes these program units stored in the memory to achieve the corresponding functions. All of the above modules are located in the same processor; alternatively, the modules may be located in different processors in any combination.

[0188] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can improve the accuracy of early warning systems for loosening mine structures.

[0189] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0190] This invention provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the diagnostic method for loosening of the mine structure.

[0191] Specifically, diagnostic methods for loosening of mine structures include:

[0192] Step S201: Acquire multi-source monitoring data and ventilation record data within the ventilation flow field. The multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data.

[0193] Step S202: Preprocess the multi-source monitoring data and merge the multi-source monitoring data and ventilation record data into fused monitoring data. The preprocessing includes feature extraction, spatiotemporal alignment and noise reduction.

[0194] Step S203: Calculate the anomaly degree, trend score and consistency index of the fused monitoring data. The anomaly degree is determined by statistical methods or machine learning methods, and the trend score is determined by a time series processing model, which includes at least a long short-term memory model.

[0195] Step S204: Weight the anomaly score, trend score, and consistency index to obtain the loosening trend score;

[0196] Step S205: If the loosening trend score is greater than or equal to the first threshold, generate an alarm message.

[0197] This invention provides a processor for running a program, wherein the program executes the diagnostic method for loosening of the mine structure.

[0198] Specifically, diagnostic methods for loosening of mine structures include:

[0199] Step S201: Acquire multi-source monitoring data and ventilation record data within the ventilation flow field. The multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data.

[0200] Step S202: Preprocess the multi-source monitoring data and merge the multi-source monitoring data and ventilation record data into fused monitoring data. The preprocessing includes feature extraction, spatiotemporal alignment and noise reduction.

[0201] Step S203: Calculate the anomaly degree, trend score and consistency index of the fused monitoring data. The anomaly degree is determined by statistical methods or machine learning methods, and the trend score is determined by a time series processing model, which includes at least a long short-term memory model.

[0202] Step S204: Weight the anomaly score, trend score, and consistency index to obtain the loosening trend score;

[0203] Step S205: If the loosening trend score is greater than or equal to the first threshold, generate an alarm message.

[0204] This invention provides a security monitoring system, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs at least the following steps:

[0205] Step S201: Acquire multi-source monitoring data and ventilation record data within the ventilation flow field. The multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data.

[0206] Step S202: Preprocess the multi-source monitoring data and merge the multi-source monitoring data and ventilation record data into fused monitoring data. The preprocessing includes feature extraction, spatiotemporal alignment and noise reduction.

[0207] Step S203: Calculate the anomaly degree, trend score and consistency index of the fused monitoring data. The anomaly degree is determined by statistical methods or machine learning methods, and the trend score is determined by a time series processing model, which includes at least a long short-term memory model.

[0208] Step S204: Weight the anomaly score, trend score, and consistency index to obtain the loosening trend score;

[0209] Step S205: If the loosening trend score is greater than or equal to the first threshold, generate an alarm message.

[0210] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having at least the following method steps:

[0211] Step S201: Acquire multi-source monitoring data and ventilation record data within the ventilation flow field. The multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data.

[0212] Step S202: Preprocess the multi-source monitoring data and merge the multi-source monitoring data and ventilation record data into fused monitoring data. The preprocessing includes feature extraction, spatiotemporal alignment and noise reduction.

[0213] Step S203: Calculate the anomaly degree, trend score and consistency index of the fused monitoring data. The anomaly degree is determined by statistical methods or machine learning methods, and the trend score is determined by a time series processing model, which includes at least a long short-term memory model.

[0214] Step S204: Weight the anomaly score, trend score, and consistency index to obtain the loosening trend score;

[0215] Step S205: If the loosening trend score is greater than or equal to the first threshold, generate an alarm message.

[0216] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0217] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0218] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0219] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0220] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0221] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0222] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0223] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0224] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0225] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0226] As can be seen from the above description, the embodiments of this application achieve the following technical effects:

[0227] 1) The method for diagnosing mine structural loosening in this application firstly acquires multi-source monitoring data and ventilation record data within the ventilation flow field. The multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data. Then, the multi-source monitoring data is preprocessed, and the multi-source monitoring data and ventilation record data are fused into fused monitoring data. The preprocessing includes feature extraction, spatiotemporal alignment, and noise reduction. Afterward, the anomaly degree, trend score, and consistency index of the fused monitoring data are calculated. The anomaly degree is determined using statistical or machine learning methods, and the trend score is determined using a time-series processing model, which includes at least a long short-term memory model. Then, the anomaly degree, trend score, and consistency index are weighted and calculated to obtain a loosening trend score. Finally, if the loosening trend score is greater than or equal to a first threshold, an alarm message is generated. This application ensures the comprehensiveness of the basis for loosening prediction by fusing multi-source data, and comprehensively analyzes the possibility of mine structure loosening (loosening trend score) from multiple perspectives, including spatial (anomaly degree), temporal (trend score) and data integrity (consistency), so as to provide early warning of the risk of mine structure loosening and improve the accuracy of early warning.

[0228] 2) The mine structure loosening diagnostic device of this application comprises: a first acquisition unit acquiring multi-source monitoring data and ventilation record data within the ventilation flow field, the multi-source monitoring data including wind speed monitoring data, sound disturbance data, and fiber optic sensing data; a first processing unit preprocessing the multi-source monitoring data and fusing the multi-source monitoring data and ventilation record data into fused monitoring data, the preprocessing including feature extraction, spatiotemporal alignment, and noise reduction; a first calculation unit calculating the anomaly degree, trend score, and consistency index of the fused monitoring data, the anomaly degree being determined using statistical or machine learning methods, and the trend score being determined using a time-series processing model, which includes at least a long short-term memory model; a second calculation unit performing weighted calculations on the anomaly degree, trend score, and consistency index to obtain a loosening trend score; and a generation unit generating an alarm message when the loosening trend score is greater than or equal to a first threshold. This application ensures the comprehensiveness of the basis for loosening prediction by fusing multi-source data, and comprehensively analyzes the possibility of mine structure loosening (loosening trend score) from multiple perspectives, including spatial (anomaly degree), temporal (trend score) and data integrity (consistency), so as to provide early warning of the risk of mine structure loosening and improve the accuracy of early warning.

[0229] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for diagnosing loosening of mine structures, characterized in that, include: Acquire multi-source monitoring data and ventilation record data within the ventilation flow field, wherein the multi-source monitoring data includes wind speed monitoring data, sound disturbance data, and fiber optic sensing data; The multi-source monitoring data is preprocessed, and the multi-source monitoring data and the ventilation record data are fused into fused monitoring data. The preprocessing includes feature extraction, spatiotemporal alignment, and noise reduction. The anomaly degree, trend score, and consistency index of the fused monitoring data are calculated. The anomaly degree is determined by statistical or machine learning methods, and the trend score is determined by a time series processing model, which includes at least a long short-term memory model. The loosening trend score is obtained by weighting the anomaly degree, the trend score, and the consistency index. An alarm message is generated if the loosening trend score is greater than or equal to a first threshold.

2. The method according to claim 1, characterized in that, Acquire multi-source monitoring data and ventilation record data within the ventilation flow field, including: The system receives monitoring signals from a first wind speed sensor group and a second wind speed sensor group to obtain the wind speed monitoring data. The first wind speed sensor group is deployed at the monitoring section, and the second wind speed sensor group is deployed along the roadway axis. The target acoustic signal is acquired by a sound pickup device to obtain sound disturbance data. The target acoustic signal is a broadband acoustic signal generated by ventilation airflow passing through structural gaps, abrupt cross sections and rough surfaces. The sound pickup device is deployed within a first preset range of the monitoring cross section. Vibration and temperature signals are acquired through distributed optical fibers to obtain optical fiber sensing data. The distributed optical fibers are continuously deployed along one or more surfaces of the surrounding rock, support structure, anchor cable and steel strip. The ventilation record data is obtained by acquiring fan start / stop signals, damper opening / closing status, tunneling machine and transportation equipment operating status, and personnel working hours through the data interface.

3. The method according to claim 1, characterized in that, The multi-source monitoring data is preprocessed, including: The wind speed monitoring data is bandpass filtered based on the first bandwidth, and the wind speed monitoring data is smoothed using the moving average method. The mean, variance, and turbulence intensity are calculated based on the wind speed monitoring data, and the spectral features of the wind speed monitoring data are extracted using Fourier transform to obtain the wind speed signal features. The sound disturbance data is bandpass filtered based on the second bandwidth, and the sound source of the sound disturbance data is determined by array beamforming or time difference localization. The sound disturbance data is then denoised based on the sound source. The sound disturbance data is processed using short-time Fourier transform, and the Mel-frequency cepstral coefficients, spectral centroid, bandwidth, spectral kurtosis, and amplitude gradient of the sound disturbance data are extracted to obtain acoustic signal characteristics. The fiber optic sensing data is sequentially processed by sliding window smoothing, temperature compensation, and baseline drift correction. The fiber optic sensing data is decomposed using wavelet transform, and the energy distribution and pulse events of the fiber optic sensing data are extracted to obtain the fiber optic perturbation characteristics. Spatiotemporal alignment is performed on the wind speed signal features, the acoustic signal features, and the fiber optic disturbance features.

4. The method according to claim 1, characterized in that, The multi-source monitoring data and the ventilation record data are fused into fused monitoring data, including: The weights of the multi-source monitoring data and the ventilation record data are determined by using a gradient boosting tree model to obtain the first target weights. The multi-source monitoring data and the ventilation record data are weighted and fused based on the first target weight to obtain the fused monitoring data; or A deep multimodal fusion network is used to stitch together and fuse the multi-source monitoring data and the ventilation record data to obtain the fused monitoring data.

5. The method according to claim 1, characterized in that, After preprocessing the multi-source monitoring data, the method further includes: In the event that any of the multi-source monitoring data is missing, the non-missing multi-source monitoring data and the ventilation record data are respectively input into the corresponding sub-discriminator to calculate the corresponding trend score, thereby obtaining the trend sub-score corresponding to each of the multi-source monitoring data and the ventilation record data. The sub-discriminator is obtained by training the time series processing model on the dataset of the corresponding category. In the event that the ventilation record data is missing, each of the multi-source monitoring data is input into the corresponding sub-discriminator to calculate the corresponding trend score and obtain the corresponding trend sub-score; The trend score of the fused monitoring data is obtained by processing the trend sub-score using weighted average, voting, or Bayesian fusion.

6. The method according to claim 1, characterized in that, Calculating the anomaly degree of the fused monitoring data includes: Using Hotelling's T 2 The control chart processes the multi-source monitoring data to obtain the anomaly degree; or The anomaly score is obtained by processing the multi-source monitoring data using the Isolation Forest algorithm; or The anomaly degree is obtained by processing the multi-source monitoring data using a support vector machine algorithm.

7. The method according to claim 1, characterized in that, Calculating the trend score of the fused monitoring data includes: The long short-term memory network is used to model the changing trend of the fused monitoring data within a first preset time period in the future, to obtain a predicted trend sequence; or A gated loop unit is used to model the change trend of the fused monitoring data within a first preset time period in the future to obtain the predicted trend sequence; The trend score corresponding to the fusion monitoring data is determined based on the predicted trend sequence.

8. The method according to claim 1, characterized in that, Calculating the consistency index of the fused monitoring data includes: Calculate the Pearson correlation coefficient of the fused monitoring data to obtain the consistency index; or The consistency index is obtained by calculating the dynamic time-normalized similarity of the fused monitoring data. The consistency rate of the window symbol in the fused monitoring data is calculated to obtain the consistency index.

9. The method according to claim 1, characterized in that, After acquiring ventilation record data, the method further includes: If the ventilation record data remains unchanged for a second preset duration, based on the multi-source monitoring data within a third preset duration prior to the current moment, the third preset duration is less than the second preset duration; A data baseline model is constructed based on the multi-source monitoring data, and the data baseline model includes the mean, standard deviation and quantile of each of the multi-source monitoring data. If the ventilation record data changes, the parameters in the baseline model are updated according to a preset time decay coefficient.

10. The method according to claim 1, characterized in that, After acquiring multi-source monitoring data and ventilation record data within the ventilation flow field, the method further includes: If the confidence level of the wind speed monitoring data is less than a preset value, radar point cloud data and flow field trajectory data are acquired. The flow field trajectory data is obtained by monitoring the particle trajectory of the tracer under the action of airflow. The instantaneous velocity vector field of different regions of the cross section is calculated based on the radar point cloud data and the flow field trajectory data. The vortex area, backflow intensity and velocity gradient peak value are determined based on the instantaneous velocity vector field. The vortex disturbance index and velocity extraction anomaly index are determined based on the vortex area, the backflow intensity, and the peak velocity gradient. The vortex disturbance index, the velocity extraction anomaly index, the multi-source monitoring data, and the ventilation record data are preprocessed, and then fused into fused monitoring data.

Images