Mine pipe gallery information acquisition and early warning method based on 5G communication

Abstract

The invention discloses a mine pipe gallery information acquisition and early warning method based on 5G communication, and the method comprises the steps: deploying a multi-source sensor in a mine pipe gallery, and transmitting monitoring data to an edge calculation node through a 5G network; after the edge node receives and caches the data, a recursive principal component analysis algorithm is adopted to carry out dynamic dimension reduction processing, and a covariance matrix is dynamically updated through a forgetting factor; the early warning module reads dimension reduction data from the real-time data buffer area, dynamically adjusts an early warning threshold value based on principal component score distribution, and constructs a closed-loop processing flow of early warning information generation, hierarchical pushing and equipment linkage control through a workflow engine; and finally, the early warning information is sent to a terminal user through a 5G network, and equipment response is triggered. According to the method, the data transmission real-time performance, the algorithm environment adaptability and the early warning accuracy are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to a mine pipe gallery information collection and early warning method based on 5G communication, belonging to the technical field of mine safety monitoring. BACKGROUND

[0002] As the core infrastructure of mine safety production, the mine pipe gallery system undertakes key functions such as power transmission, communication transmission, ventilation management, drainage system and material transportation, and its operation state is directly related to the safety and efficiency of mine operation. In recent years, with the in-depth promotion of mine intelligentization construction, the pipe gallery monitoring technology has gradually developed from traditional manual inspection to automation and informatization.

[0003] The existing technologies mainly include the following categories:

[0004] Wired sensor network monitoring technology: mainly adopts industrial Ethernet or CAN bus architecture, and deploys temperature, gas, stress and other sensor nodes at key positions of the pipe gallery to realize point data collection and centralized monitoring. This kind of technology ensures the stability of data transmission through wired connection, and has been applied on a large scale in some mines.

[0005] Digital twin risk assessment technology: by constructing a three-dimensional digital twin model of the pipe gallery, combining finite element analysis to simulate the process of gas leakage diffusion and structural stress change, the static risk can be visualized and simulated.

[0006] Time series data analysis and early warning technology: use synthetic aperture radar interferometry (InSAR) technology to obtain time series data of ground deformation, or collect historical monitoring data of sensors, and use deep learning models such as long short-term memory network (LSTM) to model and analyze the data trend, and predict the development trend of geological or structural instability.

[0007] Multi-parameter fusion monitoring technology: build a multi-source sensor data fusion system, use convolutional neural network and data fusion algorithm to jointly analyze the pipe gallery structure health, environmental parameters and equipment state, identify abnormal working conditions and generate early warning information.

[0008] Preliminary application of 5G communication technology: some advanced research explores the application of 5G technology in mine scene, such as building a dual 5G network architecture to support the operation of open-pit mine intelligent system, or using 5G network slicing technology to realize data isolation and priority transmission of different services, which preliminarily verifies the technical advantages of 5G in bandwidth and latency.

[0009] Although the above technologies have made some progress, they still face the following core technical problems in the complex underground mine environment:

[0010] (1) Insufficient real-time and reliability of data transmission;

[0011] The traditional industrial Ethernet architecture is affected by electromagnetic interference, line loss and other factors under complex geological conditions, and the transmission rate is usually only 20-40 Mbps, and the end-to-end transmission delay is more than 100 ms. When the monitoring data amount reaches 100,000 per cycle, the data backlog is serious, and the early warning response time is generally more than 5 seconds, which cannot meet the second-level emergency disposal demand of gas leakage, fire and other sudden danger. In addition, wired network has problems such as wiring difficulty and high cost when the tunnel extends and the sensor expands.

[0012] (2) The coverage density of sensor network and the system capacity are limited;

[0013] The mine pipe gallery is long and narrow, complex in structure, and has harsh working conditions such as strong electromagnetic interference, high temperature and high humidity. The connection density of traditional wireless technologies such as Wi-Fi or ZigBee is less than 100 per km, the communication distance is short, the penetration ability is weak, and it is difficult to realize millimeter-level deformation monitoring full coverage of kilometer-level pipe gallery, and there is a monitoring blind area. At the same time, when a large number of sensors are concurrently accessed, network conflicts are serious, and the data packet loss rate increases significantly.

[0014] (3) The time and space synchronization problem of multi-source monitoring data is prominent;

[0015] The time reference of underground distributed sensor nodes is inconsistent due to clock source drift and electromagnetic interference, and the alignment accuracy of traditional software-level timestamp is only millisecond level. When multiple data are used for principal component analysis covariance matrix calculation, time deviation will introduce pseudo correlation, causing feature extraction distortion and 5-8 percentage points decrease in early warning accuracy. The existing technology does not effectively utilize the high-precision timing capability of 5G base station to solve this problem.

[0016] (4) Algorithm parameters cannot adapt to dynamic mine conditions;

[0017] The activities such as blasting and mining in the process of mine exploitation will cause periodic severe disturbance of monitoring data. The traditional recursive principal component analysis uses a fixed forgetting factor (usually 0.95-0.99), which lags in response when the data distribution mutates, and the anomaly capture rate is less than 82%; in the stable period, it is too sensitive and the false positive rate is higher than 4%. The existing technology does not encode the unique geological activity law of mine (such as microseismic monitoring data and production operation plan) as dynamic algorithm parameters, making it difficult to realize adaptive adjustment.

[0018] (5) The contradiction between weak real-time processing capability and resource constraints;

[0019] The mine pipe corridor monitoring data has the characteristics of high dimension (single node > 50 dimensions) and fast flow (peak > 10,000 pieces per second). The traditional principal component analysis batch processing mode takes more than 3.7 seconds to process 100,000 pieces of data, and the memory occupation is > 112MB. The low-power explosion-proof edge computing node in the mine is limited by low power consumption, and the computing power and memory resources are limited (usually < 8GB), and there is a lack of mechanism to dynamically trim the complexity of the algorithm according to the resource state, which is easy to cause memory overflow or task timeout. SUMMARY

[0020] The present application is to solve the problems existing in the prior art and provide a mine pipe corridor information collection and early warning method based on 5G communication.

[0021] The technical scheme adopted by the present application is:

[0022] A mine pipe corridor information collection and early warning method based on 5G communication, comprising

[0023] Step 1: Deploying multiple source sensors in the mine pipe corridor, transmitting real-time monitoring data collected by the multiple source sensors to the edge computing node through the 5G communication network;

[0024] Step 2: The edge computing node receives and caches the real-time monitoring data, performs dynamic dimension reduction processing on the cached data by using a multi-thread concurrent detection and a recursive principal component analysis algorithm, wherein the recursive principal component analysis algorithm dynamically updates the covariance matrix by introducing a forgetting factor, obtains the dimension reduction data and stores it in the real-time data buffer;

[0025] Step 3: Constructing an early warning module, the early warning module reads the dimension reduction data from the real-time data buffer, and performs: periodically triggering a monitoring task through an early warning task timer, dynamically adjusting an early warning threshold based on the principal component score distribution of the dimension reduction data, and constructing a closed-loop processing flow of early warning information generation, hierarchical pushing and device linkage control through a workflow engine;

[0026] Step 4: Sending the early warning information to the terminal user through the 5G communication network, and triggering the automatic device response at the same time.

[0027] Further, in step 1, the edge computing node receives the clock synchronization signal of the 5G communication network, timestamps the sensor data, and performs time sequence alignment on the edge computing node with microsecond level precision, and uses the time and space aligned monitoring data for covariance matrix calculation of the recursive principal component analysis algorithm in step 2.

[0028] Further, in step 2, the forgetting factor is adaptively adjusted according to the mine production operation plan and the geological activity period identified from the microseismic monitoring data, and the value range is 0.85 to 0.90 in the active geological activity period and 0.95 to 0.99 in the quiet geological activity period.

[0029] Further, in step 1, the edge computing node is a 5G edge computing node deployed in the mine pipe gallery, and the 5G edge computing node realizes local data shunting through UPF sinking, and the end-to-end transmission delay is less than 15ms.

[0030] Further, in step 2, the number of threads used in the multi-thread concurrent detection is dynamically configured according to the number of sensor nodes, and 1 detection thread is configured for every 50 sensor nodes.

[0031] Further, in step 3, the workflow engine is an Activiti workflow engine or a BPMN2.0 standard engine, and the closed-loop processing flow includes six links of early warning information generation, level determination, message pushing, device linkage, disposal feedback and log archiving.

[0032] Further, in step 3, the hierarchical pushing is divided into three levels according to the warning level: the first level warning is pushed to the on-site operation personnel through the short message and the sound-light alarm, the second level warning increases the email pushing to the team leader, and the third level warning increases the application program pushing to the mine dispatching center and triggers the automatic danger elimination plan.

[0033] Further, in step 4, the automatic device response includes starting the ventilator for forced ventilation, cutting off the power supply of the target section or closing the pipe valve.

[0034] Further, in step 2, the number of principal components reserved by the recursive principal component analysis algorithm is dynamically determined according to the cumulative variance contribution rate, and the cumulative variance contribution rate threshold is set to 95%.

[0035] Further, the multi-source sensor includes a structure deformation sensor, a gas concentration sensor, a temperature sensor, a humidity sensor and a noise sensor.

[0036] The present application has the following beneficial effects:

[0037] (1) Through the 5G communication network and the edge computing architecture, high-speed low-delay transmission of monitoring data and real-time processing on the edge side are realized, the second-level response demand of massive monitoring data in the complex environment of the mine pipe gallery is met, and the warning response time is significantly shortened compared with the traditional wired network scheme, which saves the valuable time window for emergency disposal of sudden danger.

[0038] (2) The clock synchronization signal of the 5G communication network is used to timestamp and align the data of multiple sensors, which effectively eliminates the clock drift problem caused by electromagnetic interference in the downhole environment, improves the time consistency of multi-channel monitoring data, provides an accurate covariance matrix calculation basis for the recursive principal component analysis algorithm, and thus improves the accuracy of feature extraction.

[0039] (3) Introducing a forgetting factor adaptive mechanism based on the geological activity cycle perception enables the recursive principal component analysis algorithm to dynamically adjust parameter weights according to periodic disturbances such as mine blasting and mining, to respond quickly during periods of drastic changes in data distribution, and to maintain stable calculations during periods of stability, effectively reducing the risk of false alarms and false alarms of traditional fixed parameter algorithms under complex mining conditions.

[0040] (4) The warning threshold is dynamically adjusted based on the principal component score distribution to avoid the limitation that the traditional fixed threshold cannot adapt to environmental changes; the constructed hierarchical push and closed-loop handling process realizes the accurate delivery and automated linkage of warning information, and improves the intelligence level and handling efficiency of mine safety management.

[0041] (5) By using a multi-threaded concurrent detection and dynamic configuration strategy for the number of principal components, the computing resources of edge computing nodes are adaptively scheduled according to the monitoring scale. While ensuring real-time processing capabilities, memory and CPU usage are optimized, making it suitable for the long-term stable operation of explosion-proof and low-power equipment in underground mines.

[0042] (6) It covers the monitoring of multi-dimensional parameters such as structural deformation, gas concentration, temperature, humidity, and noise, and directly links key facilities such as ventilation, power, and valves to form a fully automated solution for monitoring, early warning, and disposal. It can be directly deployed in underground mine tunnels such as coal mines and metal mines, and has clear engineering application value and industry promotion prospects. Attached Figure Description

[0043] Figure 1 This is a functional module diagram of the mine pipeline corridor information collection and early warning system.

[0044] Figure 2 This is a schematic diagram of the platform architecture for a mine utility tunnel information collection and early warning system.

[0045] Figure 3 This is a technical architecture diagram of a mine utility tunnel information collection and early warning system.

[0046] Figure 4 This is a schematic diagram of the information collection module for the mine utility tunnel.

[0047] Figure 5 This is a schematic diagram of the mine pipeline corridor early warning module.

[0048] Figure 6 This is a flowchart of the information collection and early warning system for mine utility tunnels.

[0049] Figure 7 Figure 1 is a schematic diagram of the overall interface of the mine pipe gallery information collection and early warning system.

[0050] Figure 8 Figure 4 is a rate comparison and early warning response time result of different data transmission methods.

[0051] Figure 9 Figure 6 is a time consumption and memory occupation result of the recursive principal component analysis algorithm under different data volumes.

[0052] Figure 10 Figure 9 is a test accuracy and false alarm rate comparison result of different systems.

[0053] Figure 11 Figure 11 is a test result of the average failure-free time and system automatic recovery rate of the system. DETAILED DESCRIPTION

[0054] The present application will be further described below with reference to the accompanying drawings.

[0055] The present embodiment aims at the core needs of mine pipe gallery safety production monitoring, solves the problems of "high data transmission delay (traditional Ethernet transmission rate is only 26.58-34.18 Mbps), insufficient sensor network coverage, and poor environmental adaptability" in traditional monitoring methods, and is applied in a certain metal mine pipe gallery (length about 1200 m, covering power transmission pipeline, ventilation pipeline, and material transportation pipeline).

[0056] The pipe gallery has a complex geological environment, and there are microseismic activities caused by mining operations, high humidity (average humidity 65%-70%), electromagnetic interference (equipment operation noise 75-80 dB), and other working conditions, and it is necessary to realize real-time monitoring and hierarchical early warning of multiple risks such as "structural deformation, gas leakage, and temperature anomaly", and finally achieve the performance targets of "data transmission rate ≥ 84 Mbps, early warning response time ≤ 2.5 s, early warning accuracy rate ≥ 93%, and false alarm rate ≤ 2%".

[0057] The system for realizing the method of the present application is constructed based on "multi-level architecture + 5G communication + intelligent algorithm", and is specifically divided into platform architecture and technical architecture (such as Figure 2 and Figure 3 ), and the unified function module requirements correspond to Figure 1 (Functional module requirements of the mine pipe gallery information collection and early warning system).

[0058] As shown in Figure 1 , the system is divided into an "information collection module" and an "abnormal early warning module", and the core functions and implementation of each module are as follows:

[0059] Information collection module: needs to cover "data collection rules, sensor integration, data quality control, data dynamic monitoring, data feature extraction" five functions, and through "multi-source sensor deployment (integration of structural deformation, gas, temperature sensors) + Kafka message queue (guarantee data quality) + recursive principal component analysis (realize feature extraction)" to land, ensure the real-time and integrity of the collected data.

[0060] Abnormal early warning module: needs to cover "abnormal early warning rules, abnormal early warning strategy, early warning state access, employee information management, message notification" five functions, and through "Drools rule engine (define early warning rules) + Activiti workflow engine (formulate early warning strategy) + hierarchical push system (realize message notification) + employee information database (store the contact information of workers, team leaders and dispatch center personnel)" to land, ensure the accurate reach of early warning information.

[0061] The platform architecture is divided into "infrastructure service layer, platform service layer, software service layer", and the functions, hardware deployment and corresponding relationship with Figure 2 The corresponding relationship of elements is as follows:

[0062] (11) Infrastructure service layer: deploy 2 edge computing servers (model: Huawei Atlas500Pro), 1 5G core network UPF device (Huawei UPF5000) and 3 distributed storage servers (total storage capacity 10TB). Through 5G base station (Huawei 5GNRgNodeB), realize the global signal coverage of the pipe gallery, deploy 1 micro base station every 300m in the pipe gallery, the measured signal strength is ≥-85dBm, the end-to-end transmission delay is <15ms, provide stable resource support for "platform service layer" and "software service layer".

[0063] (12) Platform service layer: including monitoring and early warning module, information collection module, intelligent inspection module, emergency command module, material management module, service management module, the implementation details of each module are as follows:

[0064] Monitoring and early warning module: deployed on edge computing server (i.e. 5G edge computing node), responsible for real-time data analysis and abnormality determination, and data acquisition through linkage with information collection module.

[0065] Information collection module: deploy Kafka message queue to realize high-throughput transmission of multi-source sensor data (throughput ≥1000 pieces / second).

[0066] Intelligent inspection module: linkage with 20 high-definition cameras (resolution 4K) in the pipe gallery, supports automatic cruising (cruising cycle 30 minutes / time), and cruising data is real-time returned to the platform service layer.

[0067] Emergency command module: deployed in the mine dispatching center server, supporting early warning information receiving, emergency command issuing, and forming a "early warning-command" closed loop with the monitoring and early warning module.

[0068] Material management module: records the transportation volume of the material pipeline in the pipe gallery (data sampling interval 1 minute), and stores the data to the distributed storage server.

[0069] Service management module: real-time monitoring of the running state of each module (fault detection period 10 seconds / time), supporting fault self-recovery (such as automatic restart after Kafka queue interruption), and ensuring stable operation of the platform service layer.

[0070] (13) Software service layer: including data analysis platform, material management platform, information management platform, and implementation:

[0071] Data analysis platform: supporting multi-source data curve display (such as structural deformation trend, gas concentration change), time granularity can be selected as "1 second / 5 seconds / 1 minute", and users can view real-time data through the interface.

[0072] Material management platform: generates daily / weekly / monthly statistical reports of pipeline transportation volume, and supports Excel format export.

[0073] Information management platform: stores sensor account (installation position, calibration time) and early warning log (save period 1 year), and users can query historical data through "sensor ID".

[0074] The technical architecture is divided into "Internet of Things device layer, 5G communication layer, data processing layer, and business function layer" from bottom to top, and the technical details of each layer and the corresponding relationship with Figure 3 The corresponding relationship of elements is as follows:

[0075] (21) Internet of Things device layer: including structural deformation sensor, gas sensor, temperature sensor, temperature sensor, humidity sensor, noise sensor, and execution device, and the specific deployment scheme is shown in Table 1.

[0076] Table 1

[0077]

[0078] Execution device includes: exhaust fan (10 units, symmetrically deployed on both sides of the pipe gallery), pipeline valve (8 units, controlling material transportation), sound and light alarm (15 units, 1 unit every 80 m), emergency lighting (20 units, 1 unit every 60 m).

[0079] (22) 5G communication layer: including 5G base station, cloud server.

[0080] Adopting "5G independent networking (SA)" mode: sensor data is transmitted to the edge computing node (local deployment form of cloud server) through the 5G base station, and the data is locally distributed through the UPF sinking, without backhaul to the core network, and the actual end-to-end transmission delay is 12-14ms. At the same time, PTPv2 / IEEE1588 clock synchronization protocol is deployed to realize microsecond-level (±5μs) time calibration, which provides a basis for subsequent "data space-time alignment".

[0081] (23) Data processing layer: including data access module, rule engine module, real-time calculation module, storage database, the implementation details are as follows:

[0082] Data access module: support MQTT protocol (sensor data access) and HTTP protocol (instruction issuing), access success rate ≥ 99.9%.

[0083] Rule engine module: using Drools rule engine, 12 pre-warning rules are predefined (such as "methane concentration > 1% VOL and lasting for 5 seconds → first-level pre-warning").

[0084] Real-time calculation module: running recursive principal component analysis algorithm to realize dynamic dimension reduction of data.

[0085] Storage database: using InfluxDB (time series data storage, such as real-time sensor data) and MySQL (structured data storage, such as pre-warning log), data writing delay < 100ms.

[0086] (24) Business function layer: when implemented, the interface consistent with Figure 7 (mining pipe gallery information collection and pre-warning system overall interface schematic diagram) is constructed: the left side is the function menu bar (information collection, dynamic monitoring, pre-warning management, emergency response, system setting, user center), and the right side is the main operation area (integrating GIS map and three-dimensional pipe gallery model, and the abnormal area is marked with color coding: first-level pre-warning → red, second-level → yellow, third-level → blue), supporting clicking the abnormal point to view details such as "sensor ID, abnormal value, duration".

[0087] As Figure 4 , the information collection module is responsible for "multi-source data collection → space-time alignment → dynamic dimension reduction → rule matching", and the specific implementation steps are as follows:

[0088] (31) Data collection and transmission: after each sensor collects data according to the preset sampling interval, it is transmitted to the Kafka message queue (the topic is named "mine_pipe_gallery_data") through the 5G communication layer, Figure 4The "Kafka listener" in the "Kafka listener" in real time listens to the topic, and immediately stores the received data to the "link blocking queue" (the queue capacity is set to 10000 to avoid data overflow). The execution cache (execution cache) batch acquires data from the link blocking queue (100 per batch), and passes it to the fixed thread pool. The number of threads is dynamically configured according to "1 thread for every 50 sensors". In this implementation scenario, 7 types of sensors are deployed (about 30 sensor nodes in total), so 1 detection thread is configured (if the number of sensors increases to 50, the number of threads will automatically increase to 2), ensuring data processing efficiency. Figure 4

[0089] (32) Data space alignment: Since the sensor deployment location is scattered (such as the structural deformation sensor in the east section of the pipe gallery and the methane sensor in the west section), if there is no time calibration, it will cause "data misplacement". In this embodiment, before the "detection executor" processes the data, the "space-time alignment link" is set: Figure 4

[0090] Clock synchronization: The 5G base station sends a PTPv2 synchronization signal to all sensors once every 10 minutes. After receiving the signal, the sensor updates the local clock to ensure that the time deviation of all sensors is less than 5μs.

[0091] Timestamp marking: When the sensor collects each piece of data, it automatically adds a "precise timestamp (format: YYYY-MM-DDHH:MM:SS.ssssss)" to the data and transmits it to the link blocking queue.

[0092] Time alignment: After the execution cache obtains data from the link blocking queue, it first groups the multi-source data according to the "timestamp", and only groups the sensor data within the "same time window (±5μs)" (such as the structural deformation, methane, and temperature data at the "t=2024-05-2010:15:30.123456" time) into a group, and then passes it to the detection executor of the fixed thread pool, to avoid errors in subsequent covariance matrix calculation due to data time difference.

[0093] (33) Recursive principal component analysis (RPCA) data dimensionality reduction.

[0094] The detection executor receives the space-time aligned data and runs the RPCA algorithm to achieve dynamic dimensionality reduction. The specific steps and parameters are as follows:

[0095] Step 1: Data standardization processing;

[0096] Standardize the multi-source data in the same time window (such as the deformation, methane concentration, and temperature data at time t) according to the formula:

[0097] ​​

[0098] where x ij is the original data of the i-th group in the j-th dimension, μ j is the mean of the j-th dimension, σ j is the standard deviation, and the dimension effect is eliminated.

[0099] Step 2: Covariance matrix calculation

[0100] The covariance matrix of the normalized data is calculated as follows:

[0101]

[0102] where C is the covariance matrix, n is the number of samples (n = 100 groups in this embodiment, i.e. 100 time window data), x i ′ is the vector of the i-th sample. is the sample mean vector, and T represents transposition.

[0103] Step 3: Eigenvalue decomposition and principal component selection

[0104] The eigenvalue decomposition of the covariance matrix is performed as follows:

[0105]

[0106] where U and Λ are the eigenvector matrix and eigenvalue matrix respectively, K is the number of selected principal components, λ i is the i-th eigenvalue.

[0107] The number of principal components is dynamically determined according to the cumulative variance contribution rate ≥ 95%, and in this implementation scenario, the cumulative variance contribution rate of the first three principal components is 96.2%, so three principal components are retained, the data dimension is reduced from 7 to 3, the calculation efficiency is improved by about 50%, and the processed data is transmitted to Figure 4 the rule system.

[0108] Step 4: Dynamic update of the covariance matrix

[0109] A forgetting factor α is introduced to recursively update the covariance matrix

[0110]

[0111] where C n is the updated covariance matrix, C o is the old covariance matrix, and the value of α is adaptively adjusted according to the geological activity period identified by the “mine production operation plan + microseismic monitoring data”.

[0112] Active geological activity period (such as mining operation peak period: 8:00-18:00 every day, microseismic frequency >5 times / hour): α takes 0.85-0.90 (0.88 in this embodiment), accelerates the forgetting speed of old data, and ensures that the detection executor can capture new anomalies in time;

[0113] Quiet geological activity period (such as production stop period: 0:00-6:00 every day, microseismic frequency <1 time / hour): α takes 0.95-0.99 (0.97 in this embodiment), retains more old data weight, and avoids accidental fluctuations from causing the rule system to misjudge.

[0114] Step 5: Dimensionality reduction data storage;

[0115] The new data (3D) after dimensionality reduction is calculated according to the following formula (4):

[0116]

[0117] Wherein, U n and Λ n are the updated eigenvector matrix and eigenvalue matrix respectively; Z n is the new data after dimensionality reduction; W n is the new projection matrix. The detection executor sends it to Figure 4 “Database” (InfluxDB), with a storage period of 7 days (data exceeding 7 days is archived to MySQL), and is simultaneously synchronized to the real-time data buffer to provide data support for the early warning module.

[0118] (34) Rule matching and anomaly detection.

[0119] The detection executor sends the data after dimensionality reduction to Figure 4 “Rule system”, and the rule system calls the 12 pre-defined rules (such as “Principal component 1 score >2.5→structural deformation anomaly” and “Principal component 2 score >2.2→methane concentration anomaly”) in the “Rule file system” for matching:

[0120] If an abnormal rule is matched: the rule system generates “abnormal information (including sensor ID, abnormal value, timestamp, and matched rule ID)”, stores it to Figure 4 “Database”, and triggers the early warning module;

[0121] If no abnormal rule is matched: the rule system only marks the data as “normal”, stores it to the database for historical trend analysis, and does not trigger the subsequent process.

[0122] (41) Early warning module implementation: the early warning module is responsible for “real-time monitoring→abnormal processing→graded early warning→device linkage”, and the process is as follows Figure 5 , and the specific implementation steps are as follows:

[0123] (411) The "early warning task timer" is deployed to trigger the monitoring task at regular intervals (trigger period set to 1 second, consistent with sensor sampling interval): the timer sends a trigger signal to the "early warning executor" every 1 second, which reads a batch of reduced dimension data (100 groups) from the real-time data buffer and queries the "abnormal information" delivered by the "information collection module" (if any) to determine whether there are any abnormalities to be processed. If there are abnormalities (warning level > 0, Figure 5 "early warning level > 0"), the early warning executor delivers the abnormal information to the Activiti workflow engine along with the reduced dimension data; if there are no abnormalities, only the real-time monitoring interface data is updated, and no subsequent process is triggered.

[0124] (412) Abnormal feature processing layer: after receiving the information delivered by the early warning executor, the Activiti workflow engine first maps the abnormal data into a "rule engine recognizable object model" (such as "abnormal object = (sensor type: methane, abnormal value: 1.8% VOL, timestamp: 2024-05-2010:15:30.123456, warning level: to be determined)"), and then dynamically queries the "task rule variables" (such as "methane first-level warning threshold = 2% VOL, second-level warning threshold = 1% VOL") to determine the warning level. The specific determination criteria are shown in Table 2 (corresponding Figure 5 to "abnormal warning strategy" in the middle):

[0125] Table 2

[0126]

[0127] (413) Device linkage and message notification: the Activiti workflow engine triggers the "automatic control device" action and the "message notification system" push according to the warning level in the "abnormal warning strategy".

[0128] Automatic control device linkage:

[0129] First-level warning (e.g., methane concentration 2.2% VOL): start all exhaust fans (speed adjusted to maximum), cut off power supply to the target section, close pipeline valves, turn on emergency lighting and camera recording.

[0130] Second-level warning (e.g., structural deformation 6mm): start the target section exhaust fan, turn on the camera for point monitoring, and trigger the audible and light alarm.

[0131] Third-level warning (e.g., humidity 92% RH): start the dehumidification device, adjust the fan speed, and do not need to cut off the power.

[0132] Message notification system (monitoring personnel, administrators, employees, system administrators) hierarchical push:

[0133] Primary warning: Push through "sound and light alarm (on-site ringing + red light flashing)" + "short message (sent to 20 on-site workers, mobile phone pre-stored in employee information database)", push delay <1 second.

[0134] Secondary warning: On the basis of primary warning, add "email push (sent to 5 team leaders, mailbox pre-configured)", push delay <2 seconds.

[0135] Tertiary warning: On the basis of secondary, add "application push (sent to 3 mine dispatching centers)", push delay <3 seconds.

[0136] (414) Warning processing closed loop.

[0137] Activiti workflow engine builds a closed loop of "warning generation → level determination → device linkage → message push → disposal feedback → log archiving".

[0138] Disposal feedback: Relevant personnel (workers / team leaders) feedback "disposal results" (such as "methane leakage has been plugged, concentration reduced to 0.3% VOL") through system APP or dispatching terminal, feedback information is returned to Activiti workflow engine in real time;

[0139] Log archiving: Workflow engine archives "warning generation time, level, linkage device, push object, disposal result, disposal person" and other information to MySQL database, archiving format is "warning ID-date-type-disposal person-result" (such as "YW20240520001-2024-05-20-methane leakage-Zhang San-already disposed"), which is convenient for subsequent tracing.

[0140] (51) System workflow is as Figure 6 , the specific steps are as follows:

[0141] Step 1: Data acquisition and reception;

[0142] Multi-source sensors collect data at a sampling interval (0.1 seconds), which is transmitted to Figure 6 "Kafka listener" through 5G communication layer, and the listener writes data to "link blocking queue" in real time (0.2 seconds), which caches data in real time to avoid data loss (total time consumption 0.3 seconds).

[0143] Step 2: Data preprocessing and thread allocation;

[0144] The "fixed thread pool" dynamically configures 1 detection thread according to the number of sensor nodes (about 30), executes the cache to batch get data from the linked blocking queue (100 per batch, 0.1 seconds), and passes it to the "detection executor" (0.1 seconds). The detection executor first performs spatio-temporal alignment on the data (0.2 seconds, creation point 1), ensuring that the data timing is consistent (total time consumption 0.4 seconds, cumulative time consumption 0.7 seconds).

[0145] Step 3: Data dimensionality reduction and rule matching;

[0146] The detection executor calls the "recursive principal component analysis" algorithm to reduce the dimensionality of the data (0.4 seconds, including covariance matrix calculation, eigenvalue decomposition, and forgetting factor adjustment). The reduced data is passed to the "information collection rule engine" and the "monitoring rule engine" (0.2 seconds). The two engines jointly match the pre-defined rules (0.2 seconds): if an anomaly is matched, an abnormal information is generated; if it is normal, only the data is stored.

[0147] Step 4: Early warning triggering and device interaction.

[0148] The abnormal information is passed to the "early warning rule engine" (0.1 seconds). The engine determines the warning level (0.2 seconds) and triggers the "automated device response" (0.3 seconds, such as starting the exhaust fan and cutting off the power). The device action feedback signal is returned to the early warning rule engine (0.1 seconds, total time consumption 0.7 seconds, cumulative time consumption 2.2 seconds).

[0149] Step 5: Message pushing and log archiving.

[0150] The early warning rule engine calls the "message pushing" module to push the warning information according to the level (0.2 seconds, SMS / email / APP synchronous push). At the same time, the early warning log is archived to the database (0.1 seconds, total time consumption 0.3 seconds, cumulative time consumption 2.5 seconds). The whole process is completed.

[0151] Example scenario: On May 20, 2024, at 10:15:30, the methane gas sensor in the east section of the pipe gallery (milepost K0+300) collects "concentration 1.8% VOL, timestamp 2024-05-20 10:15:30.123456", and synchronously collects "temperature 38°C, structural deformation 4mm". The process is as follows:

[0152] Step 1: Data is transmitted to the Kafka listener via 5G and written to the linked blocking queue (0.3 seconds);

[0153] Step 2: The fixed thread pool allocates 1 detection thread, and the executor performs spatio-temporal alignment on the 3 groups of data (confirming that they are all "10:15:30.123456±5μs") (0.4 seconds);

[0154] Step 3: Recursive Principal Component Analysis Dimension Reduction (a = 0.88, Geological Active Period), Principal Component 2 Score = 2.3 (Matching "Methane Concentration Abnormality" Rule) (0.8 seconds);

[0155] Step 4: Early Warning Rule Engine Determines Level 2 Early Warning, Triggering East Section 2 Air Exhaustors to Start, Sound and Light Alarm to Sound (0.7 seconds);

[0156] Step 5: Push SMS to 20 Workers, Email to 5 Team Leaders, and Archive Logs (0.3 seconds);

[0157] The entire process takes 2.5 seconds, and subsequent worker feedback is "Leak sealing is completed after 10 minutes", and the system updates the disposal result.

[0158] Test Verification and Effectiveness Analysis.

[0159] This embodiment was tested for 6 months in a certain metal mine pipe gallery (January- June 2024), and the test indicators, results, and corresponding relationship with the drawings are as follows:

[0160] Data transmission performance (reference Figure 8 "Rate Comparison and Early Warning Response Time Results of Different Data Transmission Methods").

[0161] Test the transmission rate of different types of data Figure 8 (a) "Data Transmission Rate Comparison"): Structural deformation data 96.45 Mbps, methane gas data 102.37 Mbps, temperature data 90.32 Mbps, multi-source data fusion transmission 84.17 Mbps, all far exceeding the traditional Ethernet system (structural deformation data 34.18 Mbps, methane gas data 27.75 Mbps, temperature data 32.26 Mbps, multi-source data fusion transmission 26.58 Mbps), meeting the "high bandwidth" demand.

[0162] Early warning response time (reference Figure 8 (b) "Early Warning Response Time Results"), test the "detection time + trigger time" of 4 types of early warning:

[0163] Gas leakage: detection time 1.21s, early warning trigger time 0.82s, total response time 2.03s;

[0164] Structural deformation overrun: detection time 1.49s, early warning trigger time 1.08s, total response time 2.57s (close to the 2.5s threshold);

[0165] Temperature anomaly: detection time 0.92s, early warning trigger time 0.67s, total response time 1.59s;

[0166] Multi-parameter correlation early warning: detection time 2.43s, early warning trigger time 1.50s, total response time 3.93s (due to the need to correlate multi-dimensional data, the time is slightly longer, but still better than the traditional system of more than 5.50s).

[0167] Sensor accuracy (Refer to Table 3 "Accuracy Test Results of Different Sensors")

[0168] Table 3. Accuracy test results of different sensors

[0169]

[0170] All sensors have an error rate of ≤2.38%, with the temperature sensor having the lowest error rate (0.57%), the structural deformation sensor having the highest error rate (2.38%), the methane gas sensor having an error rate of 2.00%, the humidity sensor having an error rate of 0.76%, the smoke concentration sensor having an error rate of 1.74%, the wind speed sensor having an error rate of 1.79%, and the noise sensor having an error rate of 1.15%. All of these meet the "accuracy requirements for monitoring mine pipeline corridors" (industry allowable error ≤3%).

[0171] Figure 9 The results show the time consumption and memory usage of the recursive principal component analysis algorithm under different data volumes.

[0172] Testing the time consumption of the RPCA algorithm and the traditional PCA algorithm under different data volumes. Figure 9 (a) Comparison of Algorithm Time and Memory Usage Figure 9 (b) "Comparison of Algorithm Memory Usage":

[0173] With 100,000 data entries: Traditional PCA took 3.78 seconds and used 112.39 MB of memory; RPCA took 1.69 seconds and used 76.24 MB of memory, representing a 55.3% reduction in time and a 32.1% reduction in memory usage.

[0174] With 500,000 data entries: Traditional PCA took 15.26 seconds and used 302.15 MB of memory; RPCA took 6.89 seconds and used 145.72 MB of memory, representing a 54.8% reduction in time and a 51.8% reduction in memory usage.

[0175] For 900,000 data entries: Traditional PCA takes 27.12 seconds and uses 508.24 MB of memory; RPCA takes 11.27 seconds and uses 207.35 MB of memory, reducing time by 58.4% and memory usage by 59.2%, thus meeting the needs of "large-scale streaming data" processing.

[0176] Figure 11 This refers to the "system's mean time between failures (MTBF) and system auto-recovery rate test results".

[0177] Mean Time Between Failures (MTBF)Figure 11 (a) "Mean Time Between Failures (MTBF) Variation": 721.36h in the first month, 715.42h in the third month, and 708.69h in the sixth month, all exceeding 700h, with stability better than the industry average (600h);

[0178] Automatic recovery rate ( Figure 11 (b) "Automatic recovery rate changes": 98.18% in the first month, 98.54% in the third month, and 99.11% in the sixth month, such as "the system automatically restarts and resumes data reception after a brief interruption of the Kafka queue" and "the sensor automatically reconnects within 10 seconds after going offline", without the need for manual intervention.

[0179] Figure 10 Table 4 shows the comparison results of test accuracy and false alarm rate of different systems and the comparison results of actual early warning accuracy and response time of different systems.

[0180] Table 4 Comparison of actual early warning accuracy and response time of different systems

[0181]

[0182] Compared with three existing systems (MFOSSMS, MPGEEEMS, IEWIMPGS):

[0183] Early warning accuracy ( Figure 10 (a) "Comparison of System Test Accuracy" and Table 4 "Early Warning Accuracy"): The accuracy of this system in gas leakage is 96.52%, structural deformation is 94.81%, temperature abnormality is 97.22%, and multi-parameter correlation is 93.45%, all of which are higher than existing systems (MFOSSMS highest 86.43%, MPGEEEMS highest 85.13%, IEWIMPGS highest 89.32%).

[0184] False alarm rate ( Figure 10 (b) "Comparison of System False Alarm Rates": When the number of processing nodes is 1, the false alarm rate of this system is 1.98% (MFOSSMS 5.08%, MPGEEEMS 4.47%, IEWIMPGS 4.72%); when the number of processing nodes is 15, the false alarm rate of this system is 1.68% (MFOSSMS 4.36%, MPGEEEMS 2.95%, IEWIMPGS 2.14%), both of which are lower than the target of "false alarm rate < 2%".

[0185] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A method for information collection and early warning in mine utility tunnels based on 5G communication, characterized in that: include Step 1: Deploy multi-source sensors within the mine tunnel and transmit the real-time monitoring data collected by the multi-source sensors to the edge computing node via a 5G communication network; Step 2: The edge computing node receives and caches the real-time monitoring data. Through multi-threaded concurrent detection, the recursive principal component analysis algorithm is used to dynamically reduce the dimensionality of the cached data. The recursive principal component analysis algorithm dynamically updates the covariance matrix by introducing a forgetting factor to obtain the dimensionality-reduced data and stores it in the real-time data buffer. Step 3: Construct an early warning module. The early warning module reads the dimensionality-reduced data from the real-time data buffer and executes the following: periodically triggers monitoring tasks through an early warning task timer, dynamically adjusts the early warning threshold based on the principal component score distribution of the dimensionality-reduced data, and constructs a closed-loop processing flow for early warning information generation, hierarchical push and device linkage control through a workflow engine. Step 4: Send the warning information to the end user via the 5G communication network, and trigger the automated equipment to respond.

2. The method for information collection and early warning of mine utility tunnels based on 5G communication as described in claim 1, characterized in that: In step 1, the edge computing node receives the clock synchronization signal from the 5G communication network, timestamps the sensor data, and performs time alignment at the edge computing node with microsecond-level precision. The spatiotemporally aligned monitoring data is then used for covariance matrix calculation in the recursive principal component analysis algorithm in step 2.

3. The method for information collection and early warning of mine utility tunnels based on 5G communication as described in claim 1, characterized in that: In step 2, the range of the forgetting factor is adaptively adjusted according to the geological activity cycle identified by the mine production operation plan and microseismic monitoring data. The value is 0.85 to 0.90 during the active geological activity period and 0.95 to 0.99 during the quiet geological activity period.

4. The method for information collection and early warning of mine tunnels based on 5G communication as described in claim 1, characterized in that: In step 1, the edge computing node is a 5G edge computing node deployed in the mine tunnel. The 5G edge computing node achieves local data offloading through UPF sinking, and the end-to-end transmission latency is less than 15ms.

5. The method for information collection and early warning of mine tunnels based on 5G communication as described in claim 1, characterized in that: In step 2, the number of threads used in the multi-threaded concurrent detection is dynamically configured according to the number of sensor nodes, with one detection thread configured for every 50 sensor nodes.

6. The method for information collection and early warning of mine tunnels based on 5G communication as described in claim 1, characterized in that: In step 3, the workflow engine is either the Activiti workflow engine or the BPMN2.0 standard engine, and the closed-loop processing flow includes six stages: early warning information generation, level determination, message push, device linkage, handling feedback, and log archiving.

7. The method for information collection and early warning of mine tunnels based on 5G communication as described in claim 1, characterized in that: In step 3, the tiered push notification is divided into three levels according to the warning level: Level 1 warnings are pushed to on-site workers via SMS and audible and visual alarms; Level 2 warnings are pushed to the team leader via email; and Level 3 warnings are pushed to the mine dispatch center via an application and trigger the automatic hazard mitigation plan.

8. The method for information collection and early warning of mine tunnels based on 5G communication as described in claim 1, characterized in that: In step 4, the automated equipment response includes starting the ventilator for forced ventilation, cutting off the power supply to the target section, or closing the pipeline valve.

9. The method for information collection and early warning of mine tunnels based on 5G communication as described in claim 1, characterized in that: In step 2, the number of principal components retained by the recursive principal component analysis algorithm is dynamically determined based on the cumulative variance contribution rate, and the cumulative variance contribution rate threshold is set to 95%.

10. The method for information collection and early warning of mine utility tunnels based on 5G communication as described in claim 1, characterized in that: The multi-source sensors include a structural deformation sensor, a gas concentration sensor, a temperature sensor, a humidity sensor, and a noise sensor.

Images