Downhole multi-source CO ultralimit area dynamic delineation method

Abstract

The invention discloses an underground multi-source CO ultralimit area dynamic delineation method, which belongs to the field of coal mine safety and comprises the following steps: acquiring underground multi-dimensional environment and working condition data in real time through a multi-source heterogeneous sensor network; identifying a current working condition type based on the data and calling a corresponding CO source intensity function and a correction coefficient; solving a convection-diffusion partial differential equation in combination with the roadway three-dimensional model and the wind speed field to obtain a spatial continuous CO concentration field; optimizing the concentration profile sequence by using a one-dimensional convolutional neural network, and outputting high-precision boundary coordinates of the over-limit region; and based on the boundary, the concentration field and the wind speed field, generating a dynamic CO cloud picture fused with the source traceability label, and performing visualization. According to the method, the technical problems of CO multi-source superposition, inaccurate boundary delineation and disjunction of monitoring and regulation in a complex underground environment are solved, and dynamic, accurate and closed-loop management of a CO over-limit region is realized.

Description

Technical Field

[0001] This invention belongs to the field of coal mine safety, and in particular relates to a method for dynamically delineating multi-source CO over-limit areas underground. Background Technology

[0002] Currently, CO monitoring in coal mines mainly relies on traditional techniques such as deploying sparse sensors at key locations like goafs and return air corners, or conducting manual inspections. This method monitors CO concentration through localized data collection, reflecting CO concentration changes in specific areas to some extent. However, due to low sensor deployment density, incomplete spatial coverage, and limited data collection frequency, it lacks global perception of CO concentration distribution in complex underground ventilation networks and dynamic operating environments. Furthermore, traditional methods have significant shortcomings in data fusion and multi-source identification, making it difficult to effectively capture the concentration field characteristics under the coupling and superposition of multiple release sources. Especially when multiple CO release sources exist simultaneously underground, such as diesel equipment operation, blasting operations, and coal spontaneous combustion, existing technologies cannot distinguish the independent contributions of each source from the superimposed concentration field, limiting the accuracy and reliability of monitoring results.

[0003] Although existing technologies can achieve basic concentration alarms, there are still many unresolved technical problems: First, they cannot achieve accurate separation and independent source tracing of CO releases from multiple sources, making it difficult to distinguish between high-risk spontaneous combustion sources and conventional operation sources, which can easily lead to false alarms or missed alarms; Second, they lack the ability to delineate high-precision dynamic boundaries, resulting in unclear boundary positioning of out-of-limit areas, making it easy for personnel and emergency resources to enter high-risk areas without knowing the situation; Third, existing monitoring systems and ventilation control systems lack an intelligent linkage mechanism, making it impossible to automatically execute closed-loop ventilation control based on real-time concentration field changes, which makes it difficult to achieve targeted treatment and rapid emergency response for hazardous sources. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for dynamically delineating downhole multi-source CO over-limit regions, comprising: A multi-source heterogeneous sensor network deployed in key areas of underground roadways is used to collect multi-dimensional environmental and working condition data in real time. Based on the multi-dimensional environment and working condition data, the working condition type of the current downhole operation scenario is identified, and the CO source intensity release function and diffusion correction coefficient corresponding to the working condition type are called from the pre-built multi-working condition correction information library. Based on the invoked CO source intensity release function and diffusion correction coefficient, combined with the three-dimensional topology model of the tunnel and the real-time three-dimensional wind speed vector field, the convection-diffusion partial differential equation is solved to obtain the spatially continuous CO concentration field distribution. The convection-diffusion partial differential equation includes a time derivative term, a convection term, a diffusion term, and a source term based on the CO source intensity release function and diffusion correction coefficient. Along the main propagation direction of the airflow, a CO concentration profile sequence is extracted from the CO concentration field distribution, and the concentration profile sequence is input into a pre-trained one-dimensional convolutional neural network boundary optimization engine to output the boundary position coordinates of the CO excess region. Based on the boundary location coordinates, CO concentration field distribution and wind speed vector field, a dynamic CO cloud map is generated and visualized in a three-dimensional tunnel model. The dynamic CO cloud map uses color gradients to represent CO concentration distribution, arrow vectors to represent local airflow direction, red dashed lines to mark the boundary of the over-limit area, and overlays CO source tracing labels. When the dynamic CO cloud map shows that the boundary of the over-limit area is less than or equal to the preset distance threshold of the personnel gathering area, it automatically sends a control command to the mine intelligent ventilation control system to achieve ventilation control linkage.

[0005] Optionally, the multidimensional environmental and operating condition data includes CO concentration value, three-dimensional wind speed vector, air temperature, relative humidity, atmospheric pressure, real-time personnel positioning coordinates, and equipment operating status signals, wherein the multi-source heterogeneous sensor network collects the data in real time at a sampling frequency of not less than 1Hz.

[0006] Optionally, the identified operating condition type is based on the equipment operating status signal and dynamic characteristics of CO concentration, wherein: If the peak vibration acceleration detected is greater than or equal to 5 meters per square second and the duration is less than or equal to 3 seconds, it is determined to be a blasting operation condition; If CO concentration and NO x If the concentration ratio is greater than or equal to 5 and the temperature rise rate is greater than or equal to 2 degrees Celsius per minute, it is determined to be a coal spontaneous combustion disaster condition. If a diesel vehicle ID signal is detected and the CO concentration shows a periodic pulse increase, it is determined to be a diesel-powered equipment operating condition. Otherwise, it is considered a normal production condition.

[0007] Optionally, before solving the convection-diffusion partial differential equation, a separation process for the multi-source CO superposition region is also included: Based on the operating condition identification results, it is determined that the number of existing CO emission sources is greater than or equal to 2. For each identified CO emission source, calculate its theoretical concentration field under undisturbed conditions individually; Construct a superposition model to superimpose the theoretical concentration fields with weights, where the weights are initially set to 1. The least squares method is used to iteratively optimize the weights based on the measured concentration field, so as to minimize the error between the superposition model and the measured concentration field. Output the separated source contribution concentration fields, and perform boundary optimization and dynamic CO cloud map generation steps for each source contribution concentration field.

[0008] Optionally, the CO source intensity release function stored in the multi-condition correction information database includes: For diesel-powered equipment operating conditions, the CO source intensity release function is based on the linear change of engine speed; For blasting operations, the CO source intensity release function is based on an exponential decay model, where the decay time constant is between 30 and 120 seconds. For coal spontaneous combustion disaster conditions, the CO source intensity release function is based on the exponential growth of local temperature. For normal production conditions, the CO source intensity release function is the equivalent source intensity corresponding to the background concentration.

[0009] Optionally, the network structure of the one-dimensional convolutional neural network boundary optimization engine includes: The input layer receives a concentration sequence of length 100, corresponding to a tunnel length of 10 meters, with a sampling interval of 0.1 meters; The first convolutional layer has 64 convolutional kernels with a kernel size of 5 and the activation function is ReLU. The second convolutional layer has 32 convolutional kernels with a kernel size of 3 and the activation function is ReLU. The flattening layer flattens the convolutional output into a one-dimensional vector. The fully connected output layer has a single neuron, and the coordinates of the output boundary position are between 0 and L, where L is the length of the concentration profile sequence.

[0010] Optionally, the one-dimensional convolutional neural network boundary optimization engine is trained using 10,000 sets of real boundary data calibrated by the intelligent ventilation network simulation experiment system. The loss function is the mean squared error. After training, the boundary prediction error on the test set has a 95% confidence interval of no more than ±0.25 meters.

[0011] Optionally, a closed-loop verification step may also be included: Receive CO concentration field data for the next cycle after the control is implemented, and calculate the volume change rate of the out-of-limit region; If the volume shrinkage rate of the over-limit area is greater than or equal to 30% per minute, the current control strategy will be maintained.

[0012] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0013] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0014] Compared with the prior art, the present invention has the following advantages and technical effects: This invention achieves high-precision, dynamic delineation and closed-loop intelligent control of multi-source CO exceeding-limit areas underground by deploying a multi-source heterogeneous sensor network and integrating intelligent algorithms. Specifically, this method utilizes multi-source data fusion and operating condition identification technology to accurately identify and separate the contribution concentrations of different release sources (such as diesel equipment, blasting, and spontaneous combustion of coal), solving the problem of accurate source tracing in multi-source superposition environments. By solving the convection-diffusion equation and combining it with a one-dimensional convolutional neural network boundary optimization engine, the accuracy of hazardous area identification is significantly improved. Based on dynamically generated CO cloud maps and their source tracing tags, the system can automatically trigger and execute linkage control with the mine's intelligent ventilation system (such as adjusting the main fan, starting and stopping local fans, and controlling air doors), and use the concentration field data of the next cycle to verify the closed-loop effect and optimize the strategy. This constructs a complete technical chain from "monitoring and early warning" to "precise source tracing" and then to "effective control," ultimately greatly improving the mine's proactive prevention and control capabilities for CO disasters and the overall level of safe production. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a block diagram of a downhole multi-source heterogeneous sensor network architecture according to an embodiment of the present invention; Figure 3 This is a block diagram of the 1D-CNN boundary optimization engine network structure according to an embodiment of the present invention. Detailed Implementation

[0016] 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.

[0017] 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.

[0018] Example 1 like Figure 1 As shown, this embodiment provides a method for dynamically delineating downhole multi-source CO over-limit areas, including: A multi-source heterogeneous sensor network deployed in key areas of underground roadways is used to collect multi-dimensional environmental and working condition data in real time. Based on the multi-dimensional environment and working condition data, the working condition type of the current downhole operation scenario is identified, and the CO source intensity release function and diffusion correction coefficient corresponding to the working condition type are called from the pre-built multi-working condition correction information library. Based on the invoked CO source intensity release function and diffusion correction coefficient, combined with the three-dimensional topology model of the tunnel and the real-time three-dimensional wind speed vector field, the convection-diffusion partial differential equation is solved to obtain the spatially continuous CO concentration field distribution. The convection-diffusion partial differential equation includes a time derivative term, a convection term, a diffusion term, and a source term based on the CO source intensity release function and diffusion correction coefficient. Along the main propagation direction of the airflow, a CO concentration profile sequence is extracted from the CO concentration field distribution, and the concentration profile sequence is input into a pre-trained one-dimensional convolutional neural network boundary optimization engine to output the boundary position coordinates of the CO excess region. Based on the boundary location coordinates, CO concentration field distribution and wind speed vector field, a dynamic CO cloud map is generated and visualized in a three-dimensional tunnel model. The dynamic CO cloud map uses color gradients to represent CO concentration distribution, arrow vectors to represent local airflow direction, red dashed lines to mark the boundary of the over-limit area, and overlays CO source tracing labels. When the dynamic CO cloud map shows that the boundary of the over-limit area is less than or equal to the preset distance threshold of the personnel gathering area, it automatically sends a control command to the mine intelligent ventilation control system to achieve ventilation control linkage.

[0019] The specific process includes: S1: Through a multi-source heterogeneous sensor network deployed in key areas of underground roadways, such as Figure 2 As shown, multi-dimensional environmental and operating condition data are collected in real time at a sampling frequency of not less than 1Hz. The multi-dimensional environmental and operating condition data includes: CO concentration value, three-dimensional wind speed vector, air temperature, relative humidity, atmospheric pressure, real-time positioning coordinates of personnel, and equipment operating status signals. S2: Based on the equipment operating status signal and CO concentration dynamic characteristics, identify the operating condition type of the current downhole operation scenario. The operating condition types include: diesel power equipment operating condition, blasting operation condition, coal spontaneous combustion disaster condition, and normal production condition. Based on the identification result, call the CO source intensity release function uniquely corresponding to the operating condition type from the pre-built multi-condition correction information database. and diffusion correction coefficient ; S3: will be called As the source term, it combines the three-dimensional topological model of the tunnel with the real-time three-dimensional wind speed vector field. Solving the following convection-diffusion partial differential equation yields a spatially continuous CO concentration field distribution. : ; in, The rate of change of CO concentration over time. This represents the CO concentration (ppm), specifically referring to the spatially continuous concentration field obtained through model solutions. The velocity vector field (m / s) represents the airflow. The gas diffusion coefficient (m) 2 / s), The Laplace operator represents the concentration field and is used to describe the spatial diffusion effect of CO concentration (i.e., diffusion from high concentration areas to low concentration areas). This is a diffusion correction factor related to operating conditions. This represents the CO source intensity release function.

[0020] S4: Along the main propagation direction of the airflow, in the concentration field The length of the cut-off is CO concentration profile sequence ,in , , The concentration profile sequence is input into a pre-trained one-dimensional convolutional neural network boundary optimization engine, which outputs the boundary coordinates of the CO excess region. The positioning ambiguity of the boundary position coordinates shall not exceed ±0.25 meters; S5: Based on the boundary position coordinates Concentration field and wind speed vector field A dynamic CO cloud map is generated. In the three-dimensional tunnel model, the CO concentration distribution is represented by a color gradient, the local airflow direction is represented by an arrow vector, and the boundary of the over-limit area is marked by a red dashed line. At the same time, a CO source tracing label is superimposed. The tracing label includes the source type, equipment ID or suspected location coordinates. The dynamic CO cloud map is then pushed to the ground dispatch center's large screen, underground mobile terminals and emergency broadcast linkage system through the mining industrial network.

[0021] Furthermore, prior to step S3, a separation process for the multi-source CO superposition region is also included: Based on the operating condition identification results, the number of existing CO emission sources is determined to be M≥2; For each identified source Calculate its theoretical concentration field under undisturbed conditions separately. ; Constructing a superposition model: The weight Initially set to 1, Indicates the first The theoretical concentration field of the release source under undisturbed conditions. This represents the total theoretical concentration field after superposition. This indicates the total number of currently identified CO emission sources; The least squares method was used to measure the concentration field. Based on the baseline, iteratively optimize the weights. , making Minimum; Output the concentration fields of each source after separation And contribute concentration fields to each source respectively. Execute steps S4 to S5 to achieve independent delineation and source tracing of multi-source over-limit areas.

[0022] In the aforementioned multi-source heterogeneous sensor network, the CO concentration sensor employs non-dispersive infrared (NDIR). The detection technology has a range of 0 to 5000 ppm and a measurement accuracy of not less than ±5%FS; the three-dimensional wind speed vector is obtained by an ultrasonic anemometer with a spatial resolution of not less than 0.1 m / s; the equipment operating status signals include: engine speed and exhaust temperature in the CAN bus data of diesel vehicles, or peak value and duration of vibration acceleration in the blasting operation trigger signal.

[0023] The logic for identifying the operating condition type is as follows: If the peak vibration acceleration is detected to be ≥5m / s 2 If the duration is ≤3s, it is determined to be a blasting operation condition; If CO concentration and NO x If the concentration ratio is ≥5 and the temperature rise rate is ≥2℃ / min, it is determined to be a coal spontaneous combustion disaster condition. If a diesel vehicle ID signal is detected and the CO concentration shows a periodic pulse increase, it is determined to be a diesel-powered equipment operating condition; otherwise, it is determined to be a normal production condition.

[0024] The CO source intensity release function stored in the multi-condition correction information database is specifically as follows: Operating conditions of diesel power equipment: RPM is the engine speed. The correlation coefficient between CO emission and rotational speed. Based on the release constant, and These are constants obtained through bench testing and calibration of specific diesel engine models, and key data can be directly obtained from the CAN bus of diesel vehicles. Blasting operation conditions: ,in, This represents the initial peak CO source strength at the moment of explosion. Represents the exponentially decaying function. This indicates that the decay time constant is between 30 seconds and 120 seconds; Coal spontaneous combustion disaster conditions: ,in This represents the real-time temperature of the spontaneous combustion zone of the coal. This represents the pre-exponential factor or reaction rate constant. Indicates the temperature sensitivity coefficient. The larger the value, the faster the CO release increases due to rising temperature. This model is a simplified form of a typical Arrhenius equation, which accurately captures the core characteristic of the exponential increase in the oxidation reaction rate (manifested as CO release) with temperature during the spontaneous combustion of coal.

[0025] Normal production conditions ,in, This indicates a strong background source; physically, it means that under normal production conditions, there is no major, concentrated source of CO release. At this time, trace amounts of distributed CO release (e.g., trace amounts of desorption from the coal face, respiration by personnel, trace emissions from other equipment, etc.) together constitute a stable background concentration.

[0026] The network structure of the one-dimensional convolutional neural network boundary optimization engine includes: Input layer: Receives a concentration sequence of length 100, corresponding to a tunnel length of 10 meters, with a sampling interval of 0.1 meters; First convolutional layer: 64 convolutional kernels, kernel size 5, activation function is ReLU; The second convolutional layer has 32 convolutional kernels, each with a kernel size of 3, and uses ReLU as the activation function. Flattening layer: Flattens the convolution output into a one-dimensional vector; Fully connected output layer: single neuron, output boundary position coordinates ; The network was trained using over 10,000 sets of real boundary data calibrated by an intelligent ventilation network simulation experimental system, with the mean squared error (MSE) loss function used. After training, the boundary prediction error on the test set has a 95% confidence interval of no more than +0.25 meters.

[0027] The generation of the dynamic CO cloud map satisfies the following visualization rules: CO concentration <24ppm: Green; 24ppm≤CO concentration<50ppm: Yellow; CO concentration ≥ 50 ppm: Red; The boundary of the over-limit area is drawn with a red dashed line, 2 pixels wide, and its position is... ±0.25m; CO propagation path prediction lines are generated based on forward particle tracking of wind speed vector fields, and their length is 5 to 10 meters. The source tracing label is displayed as a floating window, containing either "Source Type: Diesel Vehicle #205" or "Suspected Spontaneous Combustion Point: Coordinates (X=125.3, Y=88.7, Z=-320.1)". Information such as "etc."

[0028] It also includes ventilation control linkage steps: When the dynamic CO cloud map shows that the boundary of the over-limit area is ≤50 meters away from the personnel gathering area, it automatically sends a control command to the mine's intelligent ventilation control system. The control commands include: Increase the frequency of the main fan upstream of the over-limit area to increase the dilution air volume; Close the air doors leading to the personnel refuge chamber to prevent CO intrusion; Start the local forced-flow fan to create a directional airflow that guides CO to be discharged into the return airway; The control effect is observed through the CO concentration field in the next cycle. Conduct closed-loop verification. If the volume reduction rate of the out-of-limit area is ≥30% / min, maintain the current control strategy; otherwise, trigger the level 2 emergency plan.

[0029] Example 2 This embodiment provides a method for dynamically delineating downhole multi-source CO over-limit areas, including: In practical applications, the first step is to deploy a multi-source heterogeneous sensor network. This network consists of various devices, including mine-use infrared CO concentration sensors, ultrasonic anemometers, temperature and humidity sensors, barometers, personnel positioning devices, and equipment operation status signal acquisition modules. These devices are installed in key areas of the roadway, working face, return air corners, and equipment passage areas to ensure coverage of the main underground operating scenarios. For example, CO concentration sensors and ultrasonic anemometers are deployed at key nodes in the roadway to monitor changes in CO concentration and wind speed vectors in the local environment in real time; temperature and humidity sensors and barometers are deployed near the working face to obtain air temperature, relative humidity, and atmospheric pressure data; personnel positioning devices and equipment operation status signal acquisition modules are deployed in equipment passage areas to record real-time personnel positioning coordinates and engine speed and exhaust temperature from diesel vehicle CAN bus data or peak and duration of vibration acceleration from blasting operation trigger signals. All sensors collect multi-dimensional environmental and operating condition data in real time at a sampling frequency of no less than 1Hz and transmit the data to underground edge computing nodes for processing.

[0030] The downhole edge computing node is one of the core components of the entire system. It integrates multiple functional modules, including a working condition identification module, a multi-working-condition correction information database storage unit, a multi-source CO separation calculation module, a strong airflow-source coupling calculation module, a 1D-CNN boundary optimization engine, a dynamic CO cloud map visualization platform, and a ventilation control command interface. The 1D-CNN boundary optimization engine, for example... Figure 3 As shown. The edge computing nodes use intrinsically safe AI servers designed for mining, with an INT8 computing power of no less than 4 TOPS, supporting real-time inference of ONNX format neural network models, and an end-to-end processing latency of no more than 800 milliseconds.

[0031] In actual operation, edge computing nodes receive data from a multi-source heterogeneous sensor network and input it into the working condition identification module for analysis. The working condition identification module determines the type of working condition for the current downhole operation based on equipment operating status signals and dynamic characteristics of CO concentration. For example, if a vibration acceleration peak ≥ 5 m / s² is detected... 2 If the duration is ≤3s, it is determined to be a blasting operation; if the CO concentration and NO concentration are ≤3s, it is determined to be a blasting operation. x If the concentration ratio is ≥5 and the temperature rise rate is ≥2℃ / min, it is determined to be a coal spontaneous combustion disaster condition; if a diesel vehicle ID signal is detected and the CO concentration shows a periodic pulse increase, it is determined to be a diesel-powered equipment operating condition; otherwise, it is determined to be a normal production condition. Based on the identification results, the condition identification module calls the CO source intensity release function uniquely corresponding to the condition type from the pre-built multi-condition correction information database storage unit. and diffusion correction coefficient .For example: Operating conditions of diesel power equipment: RPM is the engine speed; Blasting operation conditions: ,in It is directly proportional to the explosive equivalent. ; Coal spontaneous combustion disaster conditions: ,in Local temperature; Normal production worker , where is the equivalent source strength corresponding to the background concentration.

[0032] Next, the edge computing node will call As the source term, it combines the three-dimensional topological model of the tunnel with the real-time three-dimensional wind speed vector field. The data is then input into the multi-source CO separation calculation module for processing. The multi-source CO separation calculation module first determines the number of existing CO release sources M≥2 based on the operating condition identification results; for each identified source... Calculate its theoretical concentration field under undisturbed conditions separately. Next, we construct the overlay model: The weight Initially set to 1; using the least squares method, based on the measured concentration field Based on the baseline, iteratively optimize the weights. , making Minimum; output the concentration fields of each source contribution after separation and for each Execute steps S4 to S5 to achieve independent delineation and source tracing of multi-source over-limit areas.

[0033] Based on this, the wind-source strong coupling calculation module will separate the source contribution concentration fields. Input the convection-diffusion partial differential equation solver to solve the following convection-diffusion partial differential equation, obtaining the spatially continuous CO concentration field distribution. : ,in The gas diffusion coefficient is... These are the correction coefficients retrieved from the multi-condition correction information database.

[0034] To further improve the positioning accuracy of the CO over-limit region boundary, the edge computing nodes will incorporate the concentration field. The input is processed by a 1D-CNN boundary optimization engine. The 1D-CNN boundary optimization engine operates along the main propagation direction of the airflow in the concentration field. The length of the cut-off is CO concentration profile sequence ,in , , The concentration profile sequence is input into a pre-trained one-dimensional convolutional neural network boundary optimization engine, which outputs the boundary coordinates of the CO excess region. The positioning ambiguity of the boundary location coordinates does not exceed ±0.25 meters. The network structure of the 1D-CNN boundary optimization engine includes an input layer, a first convolutional layer, a second convolutional layer, a flattening layer, and a fully connected output layer.

[0035] The input layer receives a concentration sequence of length 100, corresponding to a 10-meter alleyway length, with a sampling interval of 0.1 meters. The first convolutional layer contains 64 convolutional kernels with a kernel size of 5 and the activation function is ReLU. The second convolutional layer contains 32 convolutional kernels with a kernel size of 3 and the activation function is ReLU. The flattening layer flattens the convolutional output into a one-dimensional vector. The fully connected output layer contains a single neuron, outputting the coordinates of the boundary position. ∈[0,L]. The network was trained using over 10,000 sets of real boundary data calibrated by an intelligent ventilation network simulation experimental system. The loss function used was mean squared error (MSE). After training, the boundary prediction error on the test set has a 95% confidence interval of no more than ±0.25 meters.

[0036] After completing the boundary optimization, based on the boundary position coordinates Concentration field and wind speed vector field A dynamic CO cloud map is generated. In the three-dimensional tunnel model, the CO concentration distribution is represented by a color gradient, the local airflow direction is represented by an arrow vector, and the boundary of the over-limit area is marked by a red dashed line. At the same time, a CO source tracing label is superimposed. The tracing label includes the source type, equipment ID or suspected location coordinates. The dynamic CO cloud map is then pushed to the ground dispatch center's large screen, underground mobile terminals and emergency broadcast linkage system through the mining industrial network.

[0037] The dynamic CO cloud map is generated according to the following visualization rules: CO concentration < 24 ppm is displayed in green; CO concentration 24 ppm ≤ CO concentration < 50 ppm is displayed in yellow; CO concentration ≥ 50 ppm is displayed in red; the boundary of the excess area is drawn with a red dashed line, 2 pixels wide, at the following location. ±0.25m; the CO propagation path prediction line is generated based on forward particle tracking of the wind speed vector field, with a length of 5–10 meters; the source tracing label is displayed in a floating window, containing either "Source Type: Diesel Vehicle #205" or "Suspected Spontaneous Combustion Point: Coordinates (X=125.3, Y=88.7, Z=-320.1)". Information such as "etc."

[0038] When the dynamic CO cloud map shows that the boundary of the over-limit area is ≤50 meters from the personnel gathering area, the edge computing node automatically sends control commands to the mine's intelligent ventilation control system. These commands include increasing the frequency of the main fan upstream of the over-limit area to increase the dilution air volume; closing the air doors leading to the personnel refuge chamber to prevent CO intrusion; and starting local forced-flow fans to create directional airflow to guide CO out of the return airway. The control effect is measured by the CO concentration field in the next cycle. Closed-loop verification is performed. If the volume reduction rate of the out-of-limit area is ≥30% / min, the current control strategy is maintained; otherwise, a secondary emergency plan is triggered. The ventilation control command interface is configured with a closed-loop feedback mechanism: it receives CO concentration field data for the next cycle after control execution, calculates the volume change rate of the out-of-limit area, and dynamically adjusts the fan frequency or triggers a secondary emergency plan based on whether it is ≥30% / min, thus achieving closed-loop control of "monitoring-delineation-control-verification".

[0039] In a real-world application scenario, suppose a blasting operation occurs in a coal mine, causing a rapid increase in CO concentration in a localized area. At this time, a multi-source heterogeneous sensor network collects real-time vibration acceleration peak values ​​≥5 m / s². 2Data with a duration of ≤3s is transmitted to the downhole edge computing node. The edge computing node uses the working condition identification module to determine that the current working condition is a blasting operation, and then calls the corresponding CO source intensity release function from the multi-working-condition correction information database storage unit. and diffusion correction coefficient Subsequently, the multi-source CO separation calculation module is based on the three-dimensional topology model of the tunnel and the real-time three-dimensional wind speed vector field. The spatially continuous CO concentration field distribution was calculated. The 1D-CNN boundary optimization engine further processes the concentration field data and outputs the boundary coordinates of the CO excess region. The positioning ambiguity does not exceed ±0.25 meters. A dynamic CO cloud map visualization platform generates a dynamic CO cloud map and pushes it to the ground dispatch center's large screen, underground mobile terminals, and emergency broadcast linkage system via the mining industrial network. When the dynamic CO cloud map shows that the boundary of the excess area is ≤50 meters from the personnel gathering area, the edge computing node automatically sends a control command to the mine's intelligent ventilation control system, activating local forced-flow fans to create directional airflow guiding CO to the return airway. The control effect is verified through closed-loop testing using CO concentration field data from the next cycle. If the volume reduction rate of the excess area is ≥30% / min, the current control strategy is maintained; otherwise, a level-two emergency plan is triggered.

[0040] As can be seen from the above embodiments, the method and system for dynamic delineation of underground multi-source CO over-limit areas provided in this embodiment can effectively solve the shortcomings of traditional monitoring methods in terms of global perception capability, response speed and multi-source separation, and provide technical support for mine disaster prevention and control.

[0041] To enable those skilled in the art to fully understand and implement this embodiment, the specific implementation principle of this embodiment is further explained below in conjunction with a specific application scenario.

[0042] In the underground coal mine working environment, suppose a blasting operation is underway at a working face, and diesel-powered equipment is operating in the nearby area. At this time, a multi-source heterogeneous sensor network collects real-time vibration acceleration peak values ​​≥5 m / s². 2 The data includes data with a duration of ≤3s, as well as engine speed and exhaust temperature information from the diesel vehicle's CAN bus data. This data is transmitted to the downhole edge computing node for processing via a sensor network. The edge computing node first inputs the data into the working condition identification module, determines that the current scene includes blasting operations based on vibration signal characteristics, and determines that diesel-powered equipment is also operating based on the periodic CO concentration pulse rise characteristics of the diesel vehicle.

[0043] Based on the judgment result, the operating condition identification module calls the corresponding CO source intensity release function from the multi-operating condition correction information database storage unit. and diffusion correction coefficient For blasting operations, the CO source intensity release function is called as follows: ,in The value of τ is proportional to the explosive equivalent, and its range is [30, 120] seconds; for diesel-powered equipment operating conditions, the CO source intensity release function is... Where RPM is the engine speed. Subsequently, the edge computing node will call... As the source term, it combines the three-dimensional topological model of the tunnel with the real-time three-dimensional wind speed vector field. The data is then processed by the multi-source CO separation calculation module.

[0044] The multi-source CO separation calculation module first determines the number of existing CO release sources, M=2, namely blasting operations and diesel-powered equipment operation. The module then calculates the theoretical concentration fields for both operating conditions under undisturbed conditions. And construct a superposition model The weight The initial value was set to 1. Then, the least squares method was used to measure the concentration field. Based on the baseline, iteratively optimize the weights. , making Minimum; final output is the concentration field of each source contribution after separation. .

[0045] The separated concentration field data is input into the airflow-source strength coupling calculation module to solve the convection-diffusion partial differential equations. To obtain a spatially continuous CO concentration field distribution During this process, the gas diffusion coefficient and correction factor All calculations are retrieved from the multi-condition correction information database to ensure that the calculation results can reflect the complex physical field coupling effects in the actual environment.

[0046] To further improve the positioning accuracy of the boundary of the over-limit area, the concentration field The data is fed into a 1D-CNN boundary optimization engine for processing. A CO concentration profile sequence with a length of L=10 meters is extracted from the concentration field along the main wind propagation direction. ,in , Meters. This sequence is input into a pre-trained one-dimensional convolutional neural network boundary optimization engine. After two layers of convolution operations and a fully connected layer, it outputs the boundary coordinates of the CO over-limit region. Since the network training data comes from the real boundary data calibrated by the intelligent ventilation network simulation experimental system, the positioning ambiguity of the boundary position coordinates does not exceed ±0.25 meters.

[0047] Edge computing nodes will use the boundary location coordinates Concentration field and wind speed vector field Input the data into a dynamic CO cloud map visualization platform to generate a dynamic CO cloud map. The cloud map represents the CO concentration distribution in the 3D tunnel model using a color gradient: green corresponds to <24 ppm, yellow to 24–50 ppm, and red to ≥50 ppm. The boundaries of areas exceeding the limit are marked with a 2-pixel wide red dashed line. ±0.25 meters. At the same time, the cloud map is overlaid with source tracing labels, displaying information such as "Source Type: Blasting Operation" and "Source Type: Diesel Vehicle #205".

[0048] When the dynamic CO cloud map shows that the boundary of the over-limit area is ≤50 meters from the personnel gathering area, the edge computing node automatically sends control commands to the mine's intelligent ventilation control system. These commands include increasing the frequency of the main fan upstream of the over-limit area, closing the air doors leading to the personnel refuge chamber, and starting local forced-in fans to create directional airflow to guide CO towards the return airway. The control effect is measured by the CO concentration field in the next cycle. Conduct closed-loop verification. If the volume reduction rate of the out-of-limit area is ≥30% / min, maintain the current control strategy; otherwise, trigger the level 2 emergency plan.

[0049] As can be seen from the above steps, this embodiment achieves high-precision dynamic delineation of CO concentration distribution in complex underground environments by acquiring data in real time through a multi-source heterogeneous sensor network and combining it with operating condition identification, concentration field modeling, boundary optimization, and visualization technologies. This process not only solves the shortcomings of traditional monitoring methods in terms of global perception capability, response speed, and multi-source separation, but also provides technical support for mine disaster prevention and control.

[0050] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0051] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0052] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for dynamically delineating downhole multi-source CO over-limit areas, characterized in that, include: A multi-source heterogeneous sensor network deployed in key areas of underground roadways is used to collect multi-dimensional environmental and working condition data in real time. Based on the multi-dimensional environment and working condition data, the working condition type of the current downhole operation scenario is identified, and the CO source intensity release function and diffusion correction coefficient corresponding to the working condition type are called from the pre-built multi-working condition correction information library. Based on the invoked CO source intensity release function and diffusion correction coefficient, combined with the three-dimensional topology model of the tunnel and the real-time three-dimensional wind speed vector field, the convection-diffusion partial differential equation is solved to obtain the spatially continuous CO concentration field distribution. The convection-diffusion partial differential equation includes a time derivative term, a convection term, a diffusion term, and a source term based on the CO source intensity release function and diffusion correction coefficient. Along the main propagation direction of the airflow, a CO concentration profile sequence is extracted from the CO concentration field distribution, and the concentration profile sequence is input into a pre-trained one-dimensional convolutional neural network boundary optimization engine to output the boundary position coordinates of the CO excess region. Based on the boundary location coordinates, CO concentration field distribution and wind speed vector field, a dynamic CO cloud map is generated and visualized in a three-dimensional tunnel model. The dynamic CO cloud map uses color gradients to represent CO concentration distribution, arrow vectors to represent local airflow direction, red dashed lines to mark the boundary of the over-limit area, and overlays CO source tracing labels. When the dynamic CO cloud map shows that the boundary of the over-limit area is less than or equal to the preset distance threshold of the personnel gathering area, it automatically sends a control command to the mine intelligent ventilation control system to achieve ventilation control linkage.

2. The method according to claim 1, characterized in that, The multidimensional environmental and operating condition data includes CO concentration, three-dimensional wind speed vector, air temperature, relative humidity, atmospheric pressure, real-time personnel positioning coordinates, and equipment operating status signals. The multi-source heterogeneous sensor network collects the data in real time at a sampling frequency of not less than 1Hz.

3. The method according to claim 2, characterized in that, The identification of the current downhole operation scenario's condition type is based on the equipment operating status signal and dynamic characteristics of CO concentration, wherein: If the peak vibration acceleration detected is greater than or equal to 5 meters per square second and the duration is less than or equal to 3 seconds, it is determined to be a blasting operation condition; If CO concentration and NO x If the concentration ratio is greater than or equal to 5 and the temperature rise rate is greater than or equal to 2 degrees Celsius per minute, it is determined to be a coal spontaneous combustion disaster condition. If a diesel vehicle ID signal is detected and the CO concentration shows a periodic pulse increase, it is determined to be a diesel-powered equipment operating condition. Otherwise, it is considered a normal production condition.

4. The method according to claim 1, characterized in that, Before solving the convection-diffusion partial differential equation, a separation process for the multi-source CO superposition region is also included: Based on the operating condition identification results, it is determined that the number of existing CO emission sources is greater than or equal to 2. For each identified CO emission source, calculate its theoretical concentration field under undisturbed conditions individually; Construct a superposition model to superimpose the theoretical concentration fields with weights, where the weights are initially set to 1. The least squares method is used to iteratively optimize the weights based on the measured concentration field, so as to minimize the error between the superposition model and the measured concentration field. Output the separated source contribution concentration fields, and perform boundary optimization and dynamic CO cloud map generation steps for each source contribution concentration field.

5. The method according to claim 1, characterized in that, The CO source intensity release functions stored in the multi-condition correction information database include: For diesel-powered equipment operating conditions, the CO source intensity release function is based on the linear change of engine speed; For blasting operations, the CO source intensity release function is based on an exponential decay model, where the decay time constant is between 30 and 120 seconds. For coal spontaneous combustion disaster conditions, the CO source intensity release function is based on the exponential growth of local temperature. For normal production conditions, the CO source intensity release function is the equivalent source intensity corresponding to the background concentration.

6. The method according to claim 1, characterized in that, The network structure of the one-dimensional convolutional neural network boundary optimization engine includes: The input layer receives a concentration sequence of length 100, corresponding to a tunnel length of 10 meters, with a sampling interval of 0.1 meters; The first convolutional layer has 64 convolutional kernels with a kernel size of 5 and the activation function is ReLU. The second convolutional layer has 32 convolutional kernels with a kernel size of 3 and the activation function is ReLU. The flattening layer flattens the convolutional output into a one-dimensional vector. The fully connected output layer has a single neuron, and the coordinates of the output boundary position are between 0 and L, where L is the length of the concentration profile sequence.

7. The method according to claim 6, characterized in that, The one-dimensional convolutional neural network boundary optimization engine is trained using 10,000 sets of real boundary data calibrated by the intelligent ventilation network simulation experiment system. The loss function is the mean squared error. After training, the boundary prediction error on the test set has a 95% confidence interval of no more than ±0.25 meters.

8. The method according to claim 1, characterized in that, It also includes a closed-loop verification step: Receive CO concentration field data for the next cycle after the control is implemented, and calculate the volume change rate of the out-of-limit region; If the volume shrinkage rate of the over-limit area is greater than or equal to 30% per minute, the current control strategy will be maintained.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-8.

Images