A smart monitoring and emergency control system and method for high-sulfur natural gas well sites

Abstract

This invention provides an intelligent monitoring and emergency control system and method for high-sulfur natural gas well sites, comprising a sensing and execution layer, an edge computing layer, and a cloud platform layer. The sensing and execution layer is used to collect on-site information on pipeline leakage parameters and the well site environment. The edge computing layer has a built-in control algorithm to generate pipeline control commands based on the changing trends of safety thresholds and / or on-site information, controlling the opening of electric regulating valves. The cloud platform layer communicates with the edge computing layer, constructs an adaptive gas diffusion and risk prediction model based on on-site information, and outputs probabilistic diffusion cloud maps and dynamic risk heat maps. An integrated air-space-ground monitoring network is constructed, achieving three-dimensional monitoring of the well site without blind spots through a combination of fixed-point, open-path laser, and mobile UAVs, greatly improving the timeliness of leak detection and enabling rapid, intuitive, and high-precision display of accident consequences and evolution trends.

Description

Technical Field

[0001] This invention belongs to the technical field of sulfur-containing natural gas extraction, specifically relating to an intelligent monitoring and emergency control system and method for high-sulfur natural gas well sites. Background Technology

[0002] The hydrogen sulfide (H2S) gas abundant in high-sulfur natural gas is highly toxic (safe threshold is 20 mg / m³). 3 Furthermore, it is highly corrosive to metal pipes, easily leading to thinning of pipe walls, sulfur deposition, and even leaks. In the event of a leak, it will pose an extremely serious threat to the lives and health of on-site workers, the surrounding ecological environment, and national property.

[0003] Currently, the monitoring and management of high-sulfur natural gas well sites mainly face the following technical bottlenecks: 1. Monitoring methods are limited and passive: They mainly rely on regular manual inspections and fixed-point gas sensors. Manual inspections are infrequent, risky, and difficult to perform in adverse weather or at night; fixed sensors have limited coverage and blind spots, making it impossible to achieve large-scale, three-dimensional real-time perception of the well site.

[0004] 2. Delayed Control Response: Existing control logic is mostly based on simple threshold judgments, lacking intelligent analysis and predictive control of data change trends. By the time a leak is detected, a contaminated area of ​​a certain size has often formed, resulting in a slow emergency response and missed opportunities for optimal control.

[0005] 3. Lack of systematic emergency decision support: After a leak occurs, traditional systems are unable to quickly and accurately simulate the diffusion range and path of toxic gases, and cannot provide scientific and intuitive decision-making basis for emergency evacuation of personnel and precise allocation of emergency resources.

[0006] 4. High operation and maintenance costs and low degree of automation: Frequent manual intervention, equipment maintenance, and pipeline cleaning and flow meter disassembly due to "sulfur blockage" not only increase operating costs and labor intensity, but also bring additional safety risks.

[0007] Therefore, there is a need to provide an improved technical solution that addresses the shortcomings of the existing technology. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides an intelligent monitoring and emergency control system and method for high-sulfur natural gas well sites.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A smart monitoring and emergency control system for high-sulfur natural gas well sites includes: A sensing and execution layer, deployed at the well site, is used to collect on-site information on pipeline leakage parameters and the well site environment; and The pipeline status is controlled by an electric regulating valve; An edge computing layer, which is communicatively connected to the sensing and execution layer, is used to receive information collected by the sensing and execution layer. The edge computing layer has a built-in control algorithm to generate pipeline control commands based on the changing trends of safety thresholds and / or field information, and to control the opening of the electric regulating valve. The cloud platform layer, which communicates with the edge computing layer, constructs an adaptive gas diffusion and risk prediction model based on field information and outputs a probabilistic diffusion cloud map and a dynamic risk heat map; and The predicted trend is fed forward to the edge computing layer for predictive emergency intervention.

[0010] Preferably, the perception execution layer includes: The pipeline monitoring unit includes an H2S gas concentration monitor, a pressure transmitter, a gas flow meter, and an electric regulating valve for adjusting the gas production of the pipeline installed on the pipeline. An environmental monitoring unit, comprising an infrared laser telemetry module, a miniature weather station, and video surveillance equipment.

[0011] Preferably, the perception execution layer further includes a mobile inspection unit, which is a drone equipped with a pump suction detector, a high-definition camera, an infrared thermal imager, and a positioning module.

[0012] Preferably, the control algorithms built into the edge computing layer include: The safety threshold method monitors the H2S concentration and sends a shut-off command to the electric regulating valve when the H2S concentration exceeds the threshold. The multivariable trend coordination control method comprehensively analyzes the changing trends of H2S concentration, pipeline pressure and gas flow rate, and dynamically adjusts the opening of the electric regulating valve through a fuzzy PID algorithm. Among them, the safety threshold method has a higher priority than the multivariate trend coordination control method.

[0013] Preferably, the adaptive gas diffusion and risk prediction model constructed by the cloud platform layer uses a physical information neural network as the basic predictor and couples a real-time data assimilation algorithm to form a correction engine, working collaboratively according to the following steps: Model initialization and mechanism constraint prediction: The physical information neural network performs rapid deduction of the diffusion field based on the input leakage source parameters and real-time meteorological data; the training loss function of the physical information neural network includes a data fitting term and a physical residual term. The physical residual term calculates the partial derivative of the network output with respect to the input through automatic differentiation technology and makes it satisfy the gas convection-diffusion control equation, thereby embedding the fluid dynamics mechanism as a soft constraint into the network. Real-time data assimilation and dynamic correction: The real-time data assimilation algorithm continuously receives real-time observation data sequences from the perception execution layer. It adopts a sequence assimilation method, using the model prediction field of the previous moment as the background field and the real-time observation data as the analysis field. By solving the optimal estimation problem, it dynamically generates the corrected analysis field as the optimal diffusion state estimate at the current moment. Probabilistic risk field generation and output: The uncertainty of the corrected analysis field is quantified, and a probability distribution cloud map of H2S concentration exceeding the preset threshold in the future period is generated by Monte Carlo method or ensemble forecasting technology. At the same time, personnel location information is integrated to output a dynamic personnel exposure risk heat map. Forecast trend feedforward: The heat map of personnel exposure risk is sent to the edge computing layer in real time to trigger early warning or perform pre-control operations.

[0014] Preferably, the real-time data assimilation algorithm includes continuously receiving H2S gas concentration data from the site to dynamically correct the prediction results.

[0015] Preferably, the cloud platform layer is also configured with a predictive maintenance module, which learns from historical operating data to perform fault prediction and health management.

[0016] A method for intelligent monitoring and emergency control of high-sulfur natural gas well sites, comprising well site control through any of the aforementioned intelligent monitoring and emergency control systems for high-sulfur natural gas well sites, including: Step S1: Real-time collection of pipeline leakage parameters and well site environment through the sensing and execution layer; Step S2: Perform calculation and analysis through the edge computing layer, and generate pipeline control commands based on the changing trends of safety thresholds and / or field information to control the opening degree of the electric regulating valve; Step S3: The cloud platform layer outputs a probabilistic diffusion cloud map and a dynamic risk heat map, and feeds the predicted trend forward to the edge computing layer for predictive emergency intervention.

[0017] Preferably, the edge computing layer implements a three-level alarm mechanism: The Level 1 alarm is a warning level, which is triggered by the platform when the H2S concentration exceeds the first threshold. Level 2 alarm is a dangerous level. When the H2S concentration exceeds the second threshold, it will automatically shut down the relevant pipelines and activate the on-site audible and visual alarm. A Level 3 alarm is a disaster level. When the H2S concentration exceeds the third threshold or the diffusion simulation shows a major threat, an emergency shutdown of the entire site and personnel evacuation procedures will be implemented.

[0018] Preferably, when an alarm is triggered, the cloud platform layer issues an emergency inspection task to the mobile inspection unit, and the drone goes to the designated airspace to confirm the leak point, monitor it, and transmit data back.

[0019] Beneficial effects: An integrated air-space-ground monitoring network was constructed, combining fixed-point, open-path laser, and mobile UAVs to achieve comprehensive three-dimensional monitoring of the well site without blind spots, greatly improving the timeliness of leak detection. The multivariate trend coordination control algorithm on the edge side goes beyond simple threshold judgment, enabling proactive adjustments based on dynamic data changes, ensuring both safety and efficiency, and achieving intelligent control. By integrating physical information neural networks with a data assimilation intelligent model, it can quickly, intuitively, and with high precision display the consequences and evolution trends of accidents. Attached Figure Description

[0020] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. Wherein: Figure 1 This is a schematic diagram of the overall architecture of the control system in a specific embodiment of the present invention; Figure 2 This is a flowchart of the intelligent control algorithm for the edge computing layer in a specific embodiment of the present invention; Figure 3 A schematic diagram of the structure and working mode of the mobile inspection unit in a specific embodiment provided by the invention; Figure 4 The invention provides a specific embodiment of the gas diffusion simulation effect and emergency response interface diagram under the leakage scenario; Figure 5 Examples of dynamic probabilistic risk cloud maps and personnel risk heat maps are provided in specific embodiments of the invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

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

[0023] like Figure 1-5 As shown, an intelligent monitoring and emergency control system for high-sulfur natural gas well sites includes a sensing and execution layer, an edge computing layer, and a cloud platform layer. The sensing and execution layer is deployed at the well site to collect on-site information on pipeline leakage parameters and the well site environment. Pipeline leakage parameters include H2S (hydrogen sulfide) gas concentration, pipeline pressure, and pipeline internal flow. Electric regulating valves are installed at each node of the pipeline to adjust the gas production, thereby controlling the pipeline status. The edge computing layer communicates with the sensing and execution layer and receives the collected information. The edge computing layer is not only a data aggregation node but also a local intelligent decision-making center. It has a built-in control algorithm that generates pipeline control commands based on safety thresholds and / or the changing trends of on-site information, controls the opening of the electric regulating valves, and triggers well site audible and visual alarms to issue warning information to construction personnel.

[0024] The cloud platform layer communicates with the edge computing layer. Deployed at gas gathering stations or company headquarters, the cloud platform layer is used for big data analysis, leak simulation, predictive maintenance, and emergency command. The cloud platform layer includes a data dashboard with data storage capabilities, enabling visualization and historical tracking of data across the entire well site.

[0025] Specifically, an adaptive gas diffusion and risk prediction model is constructed based on on-site information. This model is configured to receive leakage parameters and real-time meteorological data, and use an optimized time-series Gaussian plume model or computational fluid dynamics model to dynamically predict and visualize the three-dimensional diffusion range and concentration distribution of H2S gas. The output is a probabilistic diffusion cloud map and a dynamic risk heat map to guide subsequent disposal and maintenance work, forming a closed-loop management of monitoring, early warning, control and optimization.

[0026] In addition, the corrected predicted trend information can be sent to the edge computing layer in real time to dynamically adjust alarm thresholds or perform pre-control operations, thereby enabling predictive emergency intervention and improving the level of automation and intelligence. Through predictive maintenance and automated processes, it significantly reduces reliance on manual labor, maintenance costs and operational risks.

[0027] In an optional embodiment, the sensing and execution layer includes a pipeline monitoring unit and an environmental monitoring unit. The pipeline monitoring unit is used to monitor H2S gas leakage parameters of the pipeline state machine, including an H2S gas concentration monitor, a pressure transmitter, a gas flow meter, and an electric regulating valve for adjusting the gas production rate installed on the pipeline.

[0028] The environmental monitoring unit is used to monitor the on-site environment of the well site, including wind speed, wind direction, temperature, and humidity. It includes an infrared laser telemetry module (based on TDLAS technology, which can cover a path of several kilometers), a miniature weather station (monitoring wind speed, wind direction, temperature, and humidity), and video surveillance and infrared thermal imaging cameras.

[0029] The perception and execution layer also includes a mobile inspection unit, which is used to improve the collection parameters, enrich the acquisition of monitoring parameters, and build an integrated air-space-ground monitoring network, realizing three-dimensional monitoring of the well site without blind spots, and greatly improving the timeliness of leak detection.

[0030] The mobile inspection unit can be a drone equipped with a pump suction detector, a high-definition camera, an infrared thermal imager, and a positioning module. The drone has autonomous navigation, fixed-point hovering detection, and leak point tracking capabilities.

[0031] The mobile inspection unit can also be a robot dog equipped with a pump suction detector, a high-definition camera, an infrared thermal imager and a positioning module. In this embodiment, the drone or the robot dog is the mobile carrier of the mobile inspection unit, and no further restrictions are placed on it.

[0032] like Figure 1 As shown, a pipeline monitoring unit for this system is installed behind the gas production tree of each well within the well site. Data is collected via RS485 bus to the data acquisition and control unit located in the well site control room. This unit uses a Siemens S7-1500 series PLC and is equipped with a 4G / 5G communication module. Infrared laser telemetry modules are deployed around the well site, forming a cross-scanning network to monitor the H2S background concentration over a large area above the entire well site. A DJI Matrice 350 RTK drone is used as a mobile inspection unit, automatically flying twice daily along a preset route, and responding to calls to designated locations in case of anomalies.

[0033] The cloud platform is deployed at the production base 50 kilometers away, using a B / S architecture, and authorized personnel can access it in real time via a browser. The platform integrates a 3D digital twin model, intuitively displaying the actual situation at the well site.

[0034] In one optional embodiment, the edge computing layer employs an industrial-grade explosion-proof PLC or an embedded industrial control computer, which can aggregate collected information data and output control commands to the actuators (electric regulating valves) in the pipeline monitoring unit by running intelligent control algorithms. The control algorithms built into the edge computing layer include a safety threshold method and a multivariate trend coordination control method. The safety threshold method monitors the H2S concentration and sends a shut-off command to the electric regulating valve when the H2S concentration exceeds the threshold. For example, when the H2S concentration at any monitoring point exceeds a preset safety threshold (e.g., 20 mg / m³), the H2S concentration will be monitored. 3Immediately close the electric regulating valve of the corresponding well and trigger a full-field audible and visual alarm; the multivariate trend coordination control method comprehensively analyzes the changing trends of H2S concentration, pipeline pressure and gas flow, and dynamically adjusts the opening of the electric regulating valve through a fuzzy PID algorithm; for example, when the H2S concentration is rising and the pressure is stable, the opening is appropriately reduced to dilute the concentration; when the pressure drops and the flow rate increases abnormally, it is judged as a potential leakage risk, and the opening is reduced in advance.

[0035] Among them, the safety threshold method has a higher priority than the multivariate trend coordination control method, thereby ensuring the pipeline can be shut down in an emergency, thus ensuring the safety of the well site.

[0036] like Figure 2 As shown, the intelligent control algorithm running within the edge computing layer follows this process: After starting, the system cyclically collects the H2S concentration value C, pipeline pressure value P, and gas flow rate value F at each monitoring point.

[0037] Step 1: Safety Assessment. Determine if C exceeds the safety threshold S (set at 20 mg / m³). If yes, immediately send a shut-off signal to the electric regulating valve and switch to the alarm procedure. If no, proceed to the next step.

[0038] Step 2: Trend Analysis. Perform linear regression analysis on the C, P, and F data sequences of the most recent 10 sampling periods to obtain their changing trends (rising, falling, or stable).

[0039] Step 3: Multivariate coordinated decision-making.

[0040] Scenario A: If the C trend is upward and the P trend is stable or upward, then the opening of the control valve is increased in increments of 2% each time to dilute the H2S concentration by increasing the output.

[0041] Scenario B: If the P trend is decreasing and the F trend is abnormally increasing, leakage is very likely to occur. Control the valve opening to decrease rapidly in increments of 5% until the P trend stabilizes.

[0042] Scenario C: If all parameters trend steadily, then keep the current valve opening unchanged.

[0043] The cloud platform layer constructs an adaptive gas diffusion and risk prediction model based on a physical information neural network and a real-time data assimilation algorithm. The physical information neural network embeds the physical control equations of gas diffusion as constraints into the loss function. The real-time data assimilation algorithm dynamically predicts the gas diffusion based on path concentration data continuously received from the infrared laser telemetry module and point concentration data from the UAV inspection subsystem, using a time-series Gaussian plume model or a computational fluid dynamics model. It also visualizes the three-dimensional diffusion range, concentration distribution, and dynamic personnel exposure risk index of H2S gas.

[0044] The adaptive gas diffusion and risk prediction model built on the cloud platform layer uses a physical information neural network as the basic predictor and couples a real-time data assimilation algorithm to form a correction engine, working collaboratively in the following steps: Model initialization and mechanism constraint prediction: The physical information neural network uses forward propagation to quickly extrapolate the diffusion field based on the input leakage source parameters and real-time meteorological data; the training loss function of the network includes a data fitting term and a physical residual term. The physical residual term is calculated using automatic differentiation technology to determine the partial derivative of the network output with respect to the input, and makes it satisfy the simplified gas convection-diffusion control equation, thereby embedding the fluid dynamics mechanism as a soft constraint into the network; Real-time data assimilation and dynamic correction: The real-time data assimilation algorithm continuously receives real-time observation data sequences from the perception execution layer, including path integral concentration provided by the infrared laser telemetry module and point concentration provided by the UAV inspection unit; using the sequence assimilation method, the model prediction field of the previous moment is used as the background field, and the real-time observation data is used as the analysis field. By solving the optimal estimation problem, the corrected analysis field is dynamically generated as the optimal diffusion state estimate at the current moment. Probabilistic risk field generation and output: The uncertainty of the corrected analysis field is quantified, and a probability distribution cloud map of H2S concentration exceeding the preset threshold in the future period is generated by Monte Carlo method or ensemble forecasting technology. At the same time, personnel location information is integrated to output a dynamic personnel exposure risk heat map. Forecast trend feedforward: A heatmap of personnel exposure risk is sent in real time to the edge computing layer, specifically the time and probability of a diffusion front reaching critical facilities or personnel areas, to trigger early warnings or execute pre-control operations. Real-time data assimilation algorithms include continuously receiving H2S gas concentration data from the site to dynamically correct the prediction results, dynamically adjust alarm thresholds, or execute pre-control operations.

[0045] The cloud platform layer is also equipped with a predictive maintenance module. The predictive maintenance module introduces a learning model, which learns from historical operating data to perform fault prediction and health management, thereby improving the level of automation and intelligence. Through predictive maintenance and automated processes, it significantly reduces reliance on manual labor, operation and maintenance costs and operational risks.

[0046] Furthermore, a solar power supply system is installed on-site to serve as a backup power source for the sensing and execution layer and the edge computing layer, ensuring that the system's monitoring and execution functions can continue to operate even in the event of a power grid failure.

[0047] In an optional embodiment, the present invention also provides a method for intelligent monitoring and emergency control of high-sulfur natural gas well sites. Well site control is performed through any of the above-mentioned intelligent monitoring and emergency control systems for high-sulfur natural gas well sites. In step S1, pipeline leakage parameters and well site environment are collected in real time through the perception execution layer. After continuously collecting well site environment and pipeline operation data, the data is sent to the edge computing layer.

[0048] Step S2 involves performing calculations and analysis through the edge computing layer, generating pipeline control commands based on the changing trends of safety thresholds and / or field information, and controlling the opening degree of the electric regulating valve in the sensing and execution layer to regulate the flow rate of gas in the pipeline.

[0049] In step S3, the cloud platform layer synchronously receives data, constructs an adaptive gas diffusion and risk prediction model based on the on-site information, and then outputs a probabilistic diffusion cloud map and a dynamic risk heat map. The predicted trend is then fed forward to the edge computing layer for predictive emergency intervention.

[0050] In this implementation, the edge computing layer employs a three-level alarm mechanism: Level 1 alarm is a warning level, triggered by the platform when the H2S concentration exceeds a first threshold, such as when the H2S concentration exceeds 20 mg / m³. 3 The platform interface flashed and emitted a notification sound.

[0051] Level 2 alarm is a hazardous level alarm. It automatically shuts off relevant pipelines and activates on-site audible and visual alarms when the H2S concentration exceeds a second threshold, for example, when the H2S concentration exceeds 100 mg / m³. 3 It automatically shuts down the electric regulating valves of related pipelines, activates the audible and visual alarms, and automatically generates an alarm work order.

[0052] A Level 3 alarm is a disaster level. When the H2S concentration exceeds the third threshold or the diffusion simulation shows a significant threat, a full-site emergency shutdown and personnel evacuation procedures will be implemented. For example, if the H2S concentration exceeds 500 mg / m³. 3 If the model simulation shows that the high-risk area has spread to densely populated areas, an emergency shutdown will be implemented, drones will be automatically launched to verify and monitor the situation, and emergency evacuation instructions with dynamic risk heat maps will be sent to the emergency response team and surrounding personnel via SMS, App push, and other means.

[0053] In addition, the dynamic risk heat map output by the cloud platform layer identifies critically endangered areas, dangerous areas, warning areas and their uncertainty boundaries in real time, and calculates the personnel exposure risk index to provide a quantitative basis for evacuation route optimization.

[0054] like Figure 3 As shown, the system response process in the event of a pipeline leak is as follows: for example, the system triggers a level two alarm. The infrared laser telemetry module on the north side of the well site first detected that the path-average H2S concentration rose to 55 mg / m³. 3 .

[0055] After the edge computing layer confirmed the alarm, it immediately shut down the electric regulating valve of well No. 2, which was closest to the suspected area.

[0056] At the same time, an alarm window pops up on the cloud platform, and the alarm area is highlighted on the map.

[0057] The platform automatically activated the intelligent diffusion simulation and risk prediction module. Based on the data from pipeline pressure transmitter 7, the operator used the pressure difference formula to locate the leak section and estimated the leak hole diameter to be approximately 8mm. This parameter was then used in conjunction with a real-time wind speed of 3.5m / s and a southeast wind input model for initial prediction.

[0058] A few seconds later, the platform interface (such as...) Figure 4 As shown in the image, an initial leak diffusion cloud map was generated, clearly indicating that the area within 300 meters downwind is a high-risk zone (red). The commander issued a "drone emergency inspection" command with a single click through the emergency command issuance interface.

[0059] The drone automatically took off from the hangar, flew to the core leak area for close-up confirmation, and transmitted real-time video and gas concentration data back. The footage confirmed a leak at a pipe weld.

[0060] The system simultaneously sent alarm messages and evacuation maps to the mobile phones of on-site inspectors and emergency team members, instructing them to evacuate upwind.

[0061] Based on the drone's location information, the maintenance team quickly arrived at the scene to handle the situation. From the alarm to location confirmation, the entire process took no more than 8 minutes.

[0062] The cloud platform layer is also equipped with a predictive maintenance module. By training an LSTM network using data on the number of actuations, torque changes, and H2S concentration (a corrosive indicator) of the No. 1 well's electric control valve over the past year, the module predicts that the valve has an 85% probability of jamming within the next two weeks. The platform automatically generates a preventative maintenance work order and pushes it to the maintenance department. The maintenance department then inspects and replaces the valve during non-production hours, preventing an unplanned production stoppage.

[0063] Furthermore, when an alarm is triggered, the cloud platform automatically or semi-automatically issues an emergency inspection task to the mobile inspection unit. The drone goes to the designated airspace to confirm, monitor, and transmit data back to the leak point. The real-time monitoring data (such as infrared laser telemetry path concentration and drone point concentration) is dynamically assimilated into the model to achieve continuous self-correction and high-precision prediction of leak diffusion. Based on the analysis results, a graded alarm is triggered and the corresponding emergency control strategy is executed, forming a closed-loop management of monitoring-early warning-control-optimization.

[0064] In this embodiment, the adaptive gas diffusion and risk prediction model is implemented at the cloud platform layer. Its core consists of three parts: a mechanism embedding prediction module, a real-time data assimilation engine, and a risk visualization generator. The workflow follows a closed-loop logic of "offline training - online simulation - real-time correction - risk mapping". Its construction and data processing flow is as follows.

[0065] 1. Construction and training of the mechanism-embedded prediction module This module is built on a physical information neural network and aims to achieve rapid and physically consistent initial diffusion deduction.

[0066] Build process: a. Network Architecture Design: Construct a deep feedforward neural network with a high-dimensional input layer, including: leakage source features (location coordinates (x...)). s , y s The H2S prediction layer consists of source strength Q, meteorological field parameters (wind speed vector u, turbulent diffusion coefficient D), spatial grid point coordinates (x, y), and time t. The output layer is the predicted H2S concentration C(x, y, t) for the corresponding grid point.

[0067] b. Physical constraint embedding: Define the hybrid loss function L total = λ data * L data + λ phy * L phy Among them, L data This represents the mean squared error between network predictions and historical high-fidelity CFD simulations or real-world leak case data. The core innovation lies in L... phy The calculation method is as follows: Substitute the network output C(x, y, t) into the steady-state convection-diffusion control equation u·▽C - D▽ 2 C = S. The residuals of the equation are calculated using automatic differentiation techniques, and their minimization is used as the training objective. This process forces the network to learn mappings that inherently satisfy the laws of fluid dynamics.

[0068] c. Offline training: The network is trained using a large amount of historical data or high-fidelity simulation data covering different leakage scenarios (different source strengths, locations, and weather conditions). The network weights are optimized through backpropagation until the loss function converges.

[0069] 2. Data processing flow of the real-time data assimilation engine This engine is built on an ensemble Kalman filter algorithm and is responsible for fusing real-time observation data to dynamically correct mechanism predictions.

[0070] Data processing steps: a. Initializing the Prediction Set: When the system triggers an alarm (such as a level 2 alarm), the engine starts. Based on the previously trained PINN, random perturbations conforming to the prior distribution are added to its input parameters (source strength, wind speed, etc.) to generate an initial prediction set containing N members (usually N=50~100). Each member represents a possible diffusion state. The set mean X b,mean As the initial background field.

[0071] b. Observation data preprocessing and assimilation: Data input: Continuously receive two types of asynchronous time-series observation data: ① Path-averaged concentration Y from the infrared laser telemetry module path ② Point concentration Y from UAV emergency inspection point and its GPS coordinates.

[0072] Observation operator construction: Define the observation operator H to map the model state vector (grid concentration field) to the observation space. For path concentration, H is a linear integral operator along the laser path; for point concentration, H is a spatial interpolation operator.

[0073] Assimilation cycle: Executed in each assimilation period (e.g., Δt = 30s): 1. Prediction step: Each set member extrapolates a Δt forward through PINN.

[0074] 2. Update step (analysis step): Calculate the set background covariance matrix P. b Subsequently, the Kalman gain matrix K = P is calculated. b * H T * (H * P b * H T + R) -1 , where R is the observation error covariance matrix. Finally, use the formula = + Update each member to obtain the analysis set. The ensemble mean X of the analysis field a, mean This is the optimal estimate after correction based on the observed data at the current moment.

[0075] 3. Steps for generating risk visualization products The risk visualization generator receives a set of analytical fields output by the assimilation engine and generates graphical products for decision-making.

[0076] Generation steps: a. Probabilistic diffusion contour plot generation: 1. Threshold determination: For each member in the analysis set... It determines whether the concentration at each grid point exceeds a preset danger threshold (e.g., 100 mg / m³).

[0077] 2. Probability Calculation: Count the number of times the concentration exceeds the limit at each grid point out of N members, m. The probability of exceeding the limit at that point is P = m / N.

[0078] 3. Visualization Mapping: The calculated probability field P(x, y) is mapped onto an electronic map, using continuous color bands (e.g., blue-yellow-red) to represent the probability change from 0% to 100%, generating a probabilistic diffusion cloud map (e.g., ...). Figure 5 (As shown in mark 19). The cloud map is dynamically updated, reflecting the spatiotemporal evolution of the spread range and the probability of danger.

[0079] b. Generation of dynamic personnel risk heat map: 4. Personnel Location Acquisition: Connect to a personnel positioning system (such as a smart safety helmet or positioning beacon) to obtain real-time personnel coordinates (x, y, y). p , y p ).

[0080] 5. Individual Exposure Risk Calculation: For each individual, starting from the probability P of their grid point, combined with the predicted average concentration C at that point... mean A comprehensive dynamic exposure risk index R is calculated based on factors such as exposure time. p 6. Heat map creation: Centered on the location of personnel, based on their risk index R... p The size of the area is used to draw circular regions with different radii and color shades (e.g., light yellow to dark red), which are then overlaid on the electronic map to form a dynamic risk heat map of personnel (e.g., ...). Figure 5 (As shown in mark 21). High-risk individuals are highlighted and can be associated to generate optimal evacuation routes.

[0081] c. Forecast trend extraction and feedforward: From the latest probabilistic cloud map, key forecast trends are extracted, such as the estimated time when the concentration exceeds the standard in specific downwind areas (e.g., compressor stations, barracks) will reach 50%, and the movement speed of the diffusion front. This trend information is then sent to the edge computing layer in real time to trigger pre-control logic (e.g., early shutdown of equipment, adjustment of alarm thresholds).

[0082] Furthermore, the internal logic and collaborative working principle of the Adaptive Gas Diffusion and Risk Prediction Model (hereinafter referred to as the PDH model) are as follows: Figure 5 As shown, the model runs as a microservice on the cloud platform layer, and its core consists of two coupled modules: a mechanism embedding prediction module (based on PINN) and a real-time dynamic assimilation module.

[0083] The principle and collaborative working mechanism of the adaptive gas diffusion and risk prediction model are as follows: 1. Working logic of the Mechanism Embedded Prediction Module (PINN): This module is a deep feedforward neural network whose input layer receives leakage source parameters and a spatially gridded meteorological field. The network is unique in that it uses a hybrid loss function L0. total :L total = λ data * L data + λ phy * L phy Among them, L data It is the mean square error between the network output concentration field and the observed concentration fields in historical leakage cases, used to learn empirical patterns. The key innovation lies in L. phy This refers to the physical information loss term. It is calculated by substituting the concentration field C(x, t) predicted by the network into the simplified steady-state convection-diffusion equation u·▽C - D▽. 2 C = S (where u is the wind speed vector, D is the diffusion coefficient tensor, and S is the source term). Using the automatic differentiation function of PyTorch or TensorFlow, the second-order partial derivatives ∂C and ∂S of the network output with respect to spatial coordinates x and time t are directly calculated. 2 C is then used to calculate the residuals of the equation. The training objective is to minimize L. phy This means forcing the neural network's predictions to inherently satisfy the basic laws of fluid mechanics, thereby enabling physically reasonable predictions to be made even in areas with scarce data.

[0084] 2. Working logic of the real-time dynamic assimilation module: When leakage occurs, the pre-trained PINN provides the prediction field (background field X) for the first frame. b The assimilation module is initiated, its core being an Ensemble Kalman Filter (EnKF) algorithm. It maintains a state set containing dozens of members, each a prediction of PINN with slightly perturbed input parameters. This module continuously receives asynchronously arriving observation data Y. o : Path concentration from the infrared laser telemetry module: viewed as a linear operator observation H of the average concentration along a straight line. path Point concentration from UAVs: considered as observations of a specific spatial point H point .

[0085] In each assimilation cycle (e.g., 30 seconds), EnKF performs two steps: Prediction step: Each set member extrapolates forward by a short time step using PINN.

[0086] Update step (analysis step): Calculate the ensemble mean X of all member predictions. b,mean Covariance P b Then, the Kalman gain matrix K = P is calculated. b * H T * (H * P b * H T + R) -1 Where H is the observation operator and R is the observation error covariance matrix. Finally, use formula X... a = X b + K * (Y o - H * X b Update the state of each member to obtain the analysis field X. a,mean That is, the optimal estimate after correction of the observed data at the current moment.

[0087] 3. Coordination and probabilistic output of the two modules: Analysis field X after assimilation correction a,mean This serves as a more accurate initial condition for the next round of PINN short-term prediction, forming a closed loop of "prediction-assimilation-re-prediction". The probabilistic diffusion cloud map output by the model is generated by statistically analyzing the frequency with which all members of the assimilation set exceed the danger threshold at a certain location. The dynamic risk heat map of personnel is generated by spatially overlaying the probabilistic cloud map with the real-time personnel location layer to calculate the expected exposure dose at each location point.

[0088] Through the aforementioned mechanism, this model overcomes the shortcomings of traditional pure physics models, such as reliance on precise source strength and slow computation, as well as the shortcomings of pure data-driven models, such as poor extrapolation and the need for large amounts of data. It achieves a deep integration of mechanism and data, ensuring the physical rationality of predictions, real-time correction capabilities, and quantitative expression of uncertainty. Thus, it evolves the paradigm of emergency decision-making from qualitative assessment of outcomes to quantitative evaluation based on risk probability.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention shall be within the scope of protection of the pending claims of the present invention.

Claims

1. A smart monitoring and emergency control system for high-sulfur natural gas well sites, characterized in that, include: The sensing and execution layer is deployed at the well site to collect on-site information on pipeline leakage parameters and the well site environment. as well as The pipeline status is controlled by an electric regulating valve; An edge computing layer, which is communicatively connected to the sensing and execution layer, is used to receive information collected by the sensing and execution layer. The edge computing layer has a built-in control algorithm to generate pipeline control commands based on the changing trends of safety thresholds and / or field information, and to control the opening of the electric regulating valve. The cloud platform layer, which communicates with the edge computing layer, constructs an adaptive gas diffusion and risk prediction model based on field information and outputs a probabilistic diffusion cloud map and a dynamic risk heat map; and The predicted trend is fed forward to the edge computing layer for predictive emergency intervention.

2. The intelligent monitoring and emergency control system for high-sulfur natural gas well sites according to claim 1, characterized in that, The perception execution layer includes: The pipeline monitoring unit includes an H2S gas concentration monitor, a pressure transmitter, a gas flow meter, and an electric regulating valve for adjusting the gas production of the pipeline installed on the pipeline. An environmental monitoring unit, comprising an infrared laser telemetry module, a miniature weather station, and video surveillance equipment.

3. The intelligent monitoring and emergency control system for high-sulfur natural gas well sites according to claim 2, characterized in that, The perception and execution layer also includes a mobile inspection unit, which is a drone equipped with a pump suction detector, a high-definition camera, an infrared thermal imager, and a positioning module.

4. The intelligent monitoring and emergency control system for high-sulfur natural gas well sites according to claim 1, characterized in that, The control algorithms built into the edge computing layer include: The safety threshold method monitors the H2S concentration and sends a shut-off command to the electric regulating valve when the H2S concentration exceeds the threshold. The multivariable trend coordination control method comprehensively analyzes the changing trends of H2S concentration, pipeline pressure and gas flow rate, and dynamically adjusts the opening of the electric regulating valve through a fuzzy PID algorithm. Among them, the safety threshold method has a higher priority than the multivariate trend coordination control method.

5. The intelligent monitoring and emergency control system for high-sulfur natural gas well sites according to claim 1, characterized in that, The adaptive gas diffusion and risk prediction model built on the cloud platform layer uses a physical information neural network as the basic predictor and is coupled with a real-time data assimilation algorithm to form a correction engine, working collaboratively in the following steps: Model initialization and mechanism constraint prediction: The physical information neural network performs rapid deduction of the diffusion field based on the input leakage source parameters and real-time meteorological data; the training loss function of the physical information neural network includes a data fitting term and a physical residual term. The physical residual term calculates the partial derivative of the network output with respect to the input through automatic differentiation technology and makes it satisfy the gas convection-diffusion control equation, thereby embedding the fluid dynamics mechanism as a soft constraint into the network. Real-time data assimilation and dynamic correction: The real-time data assimilation algorithm continuously receives real-time observation data sequences from the perception execution layer. It adopts a sequence assimilation method, using the model prediction field of the previous moment as the background field and the real-time observation data as the analysis field. By solving the optimal estimation problem, it dynamically generates the corrected analysis field as the optimal diffusion state estimate at the current moment. Probabilistic risk field generation and output: The uncertainty of the corrected analysis field is quantified, and a probability distribution cloud map of H2S concentration exceeding the preset threshold in the future period is generated by Monte Carlo method or ensemble forecasting technology. At the same time, personnel location information is integrated to output a dynamic personnel exposure risk heat map. Forecast trend feedforward: The heat map of personnel exposure risk is sent to the edge computing layer in real time to trigger early warning or perform pre-control operations.

6. The intelligent monitoring and emergency control system for high-sulfur natural gas well sites according to claim 5, characterized in that, The real-time data assimilation algorithm includes continuously receiving H2S gas concentration data from the site to dynamically correct the prediction results.

7. The intelligent monitoring and emergency control system for high-sulfur natural gas well sites according to claim 1, characterized in that, The cloud platform layer is also equipped with a predictive maintenance module, which learns from historical operating data to perform fault prediction and health management.

8. A method for intelligent monitoring and emergency control of high-sulfur natural gas well sites, wherein well site control is performed using the intelligent monitoring and emergency control system for high-sulfur natural gas well sites as described in any one of claims 1-7, characterized in that, include: Step S1: Real-time collection of pipeline leakage parameters and well site environment through the sensing and execution layer; Step S2: Perform calculation and analysis through the edge computing layer, and generate pipeline control commands based on the changing trends of safety thresholds and / or field information to control the opening degree of the electric regulating valve; Step S3: The cloud platform layer outputs a probabilistic diffusion cloud map and a dynamic risk heat map, and feeds the predicted trend forward to the edge computing layer for predictive emergency intervention.

9. The intelligent monitoring and emergency control method for high-sulfur natural gas well sites according to claim 8, characterized in that, The edge computing layer implements a three-level alarm mechanism: The Level 1 alarm is a warning level, which is triggered by the platform when the H2S concentration exceeds the first threshold. Level 2 alarm is a dangerous level. When the H2S concentration exceeds the second threshold, it will automatically shut down the relevant pipelines and activate the on-site audible and visual alarm. A Level 3 alarm is a disaster level. When the H2S concentration exceeds the third threshold or the diffusion simulation shows a major threat, an emergency shutdown of the entire site and personnel evacuation procedures will be implemented.

10. The intelligent monitoring and emergency control method for high-sulfur natural gas well sites according to claim 8, characterized in that, When an alarm is triggered, the cloud platform layer issues an emergency inspection task to the mobile inspection unit, and the drone goes to the designated airspace to confirm the leak point, monitor it and transmit data back.

Images