Gas sensor network and hot work linkage control method

Abstract

The application discloses a kind of gas sensor network and fire linkage control method, it is related to industrial fire operation safety monitoring technical field.For the safety response lag and frequent malfunction of existing single-point threshold trigger mode, by deploying fixed and mobile gas monitoring nodes and environmental perception units, based on computational fluid dynamics and Gaussian diffusion model, combustible gas dynamic diffusion distribution field is constructed;Concentration gradient and prediction uncertainty are used to indicate mobile node patrol verification and correct the model;A reduced-order local response model is established around the fire point, and the future concentration sequence is predicted at high frequency;When the first safety threshold is exceeded, an audible and visual warning is given, and when the second safety threshold is exceeded, the power is cut off and the gas source is cut off.The present application can predict and actively interlock in advance, and is mainly used for fire operation safety control in petrochemical and other fields.

Description

Technical Field

[0001] This invention relates to the field of industrial hot work safety monitoring technology. More specifically, this invention relates to a gas sensor network and a hot work linkage control method. Background Technology

[0002] In industries such as petrochemicals, natural gas, and metallurgy, hot work (including welding, cutting, grilling, and other operations requiring open flames or high-temperature flames) is one of the most common high-risk work activities. During hot work, flammable gases may accumulate in the work area due to equipment leaks, pipeline ruptures, or operational errors. When the concentration of flammable gas reaches or exceeds its lower explosive limit (LEL), it can easily cause fires and explosions upon contact with the high temperatures or open flames generated by the hot work, resulting in serious casualties and property damage. Therefore, real-time monitoring of the flammable gas concentration in the work area during hot work and timely implementation of safety interlock actions (such as power outages, flameouts, and gas supply cutoffs) when dangerous concentrations are reached are core technical means to ensure the safety of hot work operations.

[0003] Currently, the monitoring and control of combustible gases at industrial hot work sites mainly adopts a single-point threshold triggering method. Specifically, several fixed combustible gas detectors are installed around the hot work site. Each detector operates independently, and when the real-time gas concentration at its measured location exceeds a preset fixed alarm threshold, the detector emits a local audible and visual alarm signal. Simultaneously, an interlocking action is triggered via a relay or controller to cut off the power or gas supply to the hot work equipment. For example, Chinese utility model patent CN220669198U discloses a pipeline safety monitoring device for hot work operations in oil and gas pipelines. This device integrates combustible gas detectors, flow meters, pressure transmitters, and temperature and humidity sensors. These sensors are installed on the oil and gas pipeline to monitor real-time data such as pressure, temperature, humidity, instantaneous flow rate, and combustible gas concentration. Based on the real-time data aggregation and statistics, an alarm is triggered when a preset threshold is reached. The above scheme reveals the mainstream technical approach for hot work site monitoring, which integrates multiple sensors into a single device and determines whether to trigger an alarm based on the preset concentration threshold. However, the common feature of such solutions is that, regardless of the combination of sensor types or the optimization of integration methods, their safety judgment logic is essentially still to compare the gas concentration at a "current moment and a single point" with a threshold, and trigger a response once the concentration exceeds the standard.

[0004] Single-point threshold triggering presents a core contradiction in practical industrial applications: the difficulty of balancing delayed safety response with frequent malfunctions. On one hand, because leaked gas requires a certain amount of time to diffuse and travel before reaching and being detected by a fixed detector, if the leak source is close to the ignition point while the detector is far away, the leaked gas may have already reached the ignition point before the detector detects the threshold concentration. This results in a significant delay in safety interlock actions, missing the optimal opportunity to cut off the ignition source before the gas reaches the ignition point. In industrial practice, the lack of monitoring of the dynamic correlation between temperature and concentration leads the system to rely solely on fixed thresholds for alarms, failing to anticipate parameter coupling risks and trend evolution, resulting in severely delayed warnings and ultimately accidents. On the other hand, in open or semi-open industrial environments, changes in ambient airflow can cause individual sensors to display short-term high-concentration false alarms. For example, a sudden change in wind direction may bring a small amount of leaked gas from a distance to the vicinity of a sensor. Traditional solutions cannot identify this, immediately triggering interlocks and mistakenly cutting off the ignition equipment, severely disrupting operational continuity and production efficiency. Traditional gas detection systems suffer from a false alarm rate as high as 15% to 20% due to the susceptibility of semiconductor sensors to environmental temperature and humidity interference and the aging and drift of catalytic combustion sensors. False alarms caused by sensors can lead to emergency shutdowns of the entire plant, resulting in direct economic losses exceeding one million.

[0005] Therefore, the common shortcomings of existing hot work gas monitoring and linkage control technologies are: they rely solely on fixed thresholds at single points for passive response, failing to utilize spatial correlation information between multiple sensors to determine gas diffusion trends and directions, unable to distinguish between leaks and diffusions that truly endanger the hot work site and localized concentration disturbances due to directional deviations, and unable to predict future gas concentration evolution at the hot work site based on current monitoring data and execute safety interlock actions in advance before the actual arrival of danger. Thus, the passive, reactive response mechanism can never achieve a balance between safety margin and operational continuity, which is a pressing problem to be solved in the field of hot work safety. Summary of the Invention

[0006] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0007] Another objective of this invention is to provide a gas sensor network and hot work linkage control method, which can construct a dynamic diffusion distribution field of combustible gas and combine it with a reduced-order local response model to achieve high-frequency prediction and early safety interlocking of the future concentration evolution of the hot work point, accurately distinguish the direction of dangerous diffusion from irrelevant disturbances, effectively solve the contradiction between delayed safety response and frequent malfunctions, and balance the safety and continuity of hot work operations.

[0008] To achieve these objectives and other advantages according to the present invention, a gas sensor network and hot work linkage control method is provided, comprising: A gas sensor network and environmental parameter sensing units are deployed in and around the hot work area. The gas sensor network includes fixed gas monitoring nodes deployed at high-risk locations or on key equipment and mobile gas monitoring nodes mounted on a mobile platform. Based on the real-time data collected by the gas sensor network and environmental parameter sensing unit, a dynamic diffusion distribution field of combustible gas within the work area is constructed using computational fluid dynamics and Gaussian diffusion models. Based on the predicted values ​​of combustible gas concentration at each spatial location output by the dynamic diffusion distribution field of combustible gas and the prediction uncertainty characterizing the prediction confidence, the target area indicated by the concentration gradient direction of combustible gas or the prediction uncertainty is determined. The mobile gas monitoring node is controlled to move to the target area for patrol monitoring, and the verification concentration data is obtained. Based on the deviation between the verification concentration data and the predicted concentration at the same location by the dynamic diffusion distribution field of combustible gas, the computational fluid dynamics model is corrected, and the corrected dynamic diffusion distribution field of combustible gas is obtained. Centered on the real-time location of the hot work operation, and combined with the corrected dynamic diffusion distribution field of combustible gas, a reduced-order local response model is established to predict the concentration sequence of combustible gas around the hot work point. The solution period of the reduced-order local response model is shorter than that of the computational fluid dynamics model. Within a future preset time window, if any predicted value output by the reduced-order local response model is greater than or equal to the first safety threshold, a first linkage control command is generated; if any predicted value output by the reduced-order local response model is greater than or equal to the second safety threshold, or if the ignition point concentration value output by the corrected combustible gas dynamic diffusion distribution field is greater than or equal to the second safety threshold, a second linkage control command is generated; the second safety threshold is higher than the first safety threshold. The modified computational fluid dynamics model operates at a first frequency, the reduced-order local response model operates at a second frequency, the second frequency is higher than the first frequency, and the length of the future preset time window is greater than the update cycle corresponding to the first frequency.

[0009] Preferably, the execution of the first linkage control command includes an audible and visual warning action, and the execution of the second linkage control command includes a power-off and flameout action and a gas supply cut-off action.

[0010] Preferably, the data collected in real time by the gas sensor network and the environmental parameter sensing unit includes the node location of the gas sensor network, the real-time gas concentration data of the fixed gas monitoring node, the floating detection concentration data of the mobile gas monitoring node, and the environmental wind speed and direction data.

[0011] Preferably, the construction of the dynamic diffusion distribution field of combustible gas within the work area specifically includes: The data inputs include the node locations of the gas sensor network, the real-time gas concentration data of the fixed gas monitoring nodes, the mobile gas monitoring node's floating detection concentration data, and the ambient wind speed and direction data. The environmental wind speed and direction data and the real-time gas concentration data collected by the fixed gas monitoring node are used as inputs to the Gaussian diffusion model. The location and leakage rate of the leakage source are used as unknown variables to solve the Gaussian diffusion model in reverse, and the preliminary location and leakage rate of the leakage source are derived as leakage source parameters. Using the leakage source parameters and the ambient wind speed and direction data as inputs to the Gaussian diffusion model, the initial estimated values ​​of combustible gas concentration at each spatial location within the work area are calculated to form an initial diffusion distribution field. Using the initial diffusion distribution field as the initial condition of the computational fluid dynamics model, and combining the three-dimensional geometric model and boundary conditions of the working area, the computational fluid dynamics model is used to perform numerical solutions to obtain the dynamic diffusion distribution field of the combustible gas; the dynamic diffusion distribution field of the combustible gas outputs the predicted values ​​of the combustible gas concentration at each spatial location and the prediction uncertainty characterizing the prediction confidence.

[0012] Preferably, the determination of the target region indicated by the direction of the concentration gradient of the combustible gas or the prediction uncertainty specifically includes: The predicted values ​​of combustible gas concentration at each spatial location are extracted from the dynamic diffusion distribution field of the combustible gas. The partial derivatives of the predicted values ​​of combustible gas concentration in each spatial direction are calculated to obtain the concentration gradient vector field of the combustible gas. The direction of each vector in the concentration gradient vector field points to the direction in which the combustible gas concentration rises the fastest at the corresponding location. The prediction uncertainty of each spatial location is extracted from the dynamic diffusion distribution field of the combustible gas, and the region where the prediction uncertainty exceeds a preset uncertainty threshold is identified as a low confidence region of the dynamic diffusion distribution field of the combustible gas. Based on the concentration gradient vector field and the low confidence region, the target region is determined according to priority: the region along the direction indicated by the concentration gradient vector field is taken as the first priority target region, and the part of the low confidence region of the model that does not overlap with the first priority target region is taken as the second priority target region; The mobile gas monitoring node is controlled to move to the first priority target area for patrol monitoring first, and after completing the patrol monitoring in the first priority target area, it moves to the second priority target area for patrol monitoring.

[0013] Preferably, the establishment of the reduced-order local response model specifically includes: Centered on the real-time location of the hot work operation, local concentration distribution data within a preset spatial range around the real-time location is extracted from the corrected dynamic diffusion distribution field of combustible gas and used as a local concentration snapshot sample. For the snapshot matrix composed of local concentration snapshot samples at multiple time points, an intrinsic orthogonal decomposition method is used to extract a set of dominant spatial modes, which characterize the main spatial features of the gas concentration distribution around the ignition point; Based on the dominant spatial modes, the computational fluid dynamics control equations are subjected to Galerkin projection to obtain a set of low-dimensional ordinary differential equations about the modal coefficients. The global diffusion trend output by the computational fluid dynamics model after real-time parameter assimilation and correction, when running at the first frequency, is used as the time-varying boundary condition of the low-dimensional ordinary differential equation system. The low-dimensional ordinary differential equation system is numerically integrated at the second frequency to obtain the time series of modal coefficients. The time series of modal coefficients is superimposed with the dominant spatial mode to reconstruct the gas concentration prediction sequence of the ignition point within the future preset time window.

[0014] Preferably, the alternative to the intrinsic orthogonal decomposition method is the dynamic mode decomposition method. When using the dynamic mode decomposition method, dynamic modes describing the spatiotemporal evolution of gas concentration around the ignition point and corresponding eigenvalues ​​are extracted from the snapshot matrix. A low-dimensional linear evolution model is constructed based on the dynamic modes and eigenvalues ​​to replace the Galerkin projection step and directly extrapolate to predict the gas concentration prediction sequence within the future preset time window.

[0015] Preferably, the environmental parameter sensing unit includes at least one ultrasonic anemometer; the fixed gas monitoring node and the mobile gas monitoring node employ at least one of the following sensor types: The catalytic combustion combustible gas sensor is used to detect hydrocarbon combustible gases. Its output signal has a near-linear relationship with the concentration of combustible gas and has a broad spectrum response characteristic to various combustible gases. Infrared absorption combustible gas sensors are used to detect specific combustible gases. They are based on the principle that infrared light of a specific wavelength is absorbed by the target gas molecules and are not affected by the susceptibility of catalytic combustion sensors to poisoning by sulfides or halogen compounds. Electrochemical gas sensors are used to detect the concentration of specific toxic gases or oxygen. Semiconductor-based combustible gas sensors utilize the principle of changes in the conductivity of metal oxide semiconductor materials in a combustible gas environment for detection. The fixed gas monitoring node is equipped with an explosion-proof enclosure with an explosion-proof rating of not less than ExdIICT6 and a protection rating of not less than IP65. The sampling method is either diffusion or pump suction.

[0016] The present invention has at least the following beneficial effects: Firstly, this invention establishes a hierarchical model architecture with global slow updates and local fast predictions. It utilizes a reduced-order local response model to make high-frequency predictions of the future concentration evolution at the ignition point at a frequency higher than that of the CFD model. It executes safety interlock actions in advance before the hazardous gas actually reaches the ignition point, solving the problem that the safety response lags behind the gas diffusion rate in the existing single-point threshold triggering method, and elevating safety decision-making from passive waiting to proactive prediction. Secondly, this invention constructs a concentration gradient vector field and combines it with prediction uncertainty to determine the target area of ​​the mobile gas monitoring node using a dual-criteria mechanism. It actively approaches potential leakage sources along the concentration gradient direction for verification, and performs blind detection for low-confidence areas of the model. This realizes the autonomous optimization and scheduling of the mobile node's patrol path, solving the problems of limited detection field of view and inability to self-correct model mismatch in traditional fixed-point monitoring. Thirdly, this invention uses the intrinsic orthogonal decomposition method to reduce the order of the full-order CFD model, extracts the dominant spatial modes, and establishes a low-dimensional ordinary differential equation system through Galerkin projection. While preserving the accuracy of the CFD model, it significantly reduces the solution cycle, enabling high-frequency prediction of gas concentration around the ignition point under limited computing resources. This solves the contradiction between the computation time of high-precision CFD models and the need for rapid linkage response. Fourth, the present invention, through the configuration of fixed and mobile gas monitoring nodes, can adapt to the detection needs of different combustible gases and complex industrial environments, ensuring the reliability and long-term stable operation capability of the sensor network under harsh conditions such as high temperature, high humidity, and corrosive gases.

[0017] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of the gas sensor network and hot work linkage control method described in one technical solution of the present invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0020] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0021] like Figure 1 As shown, the present invention provides a gas sensor network and hot work linkage control method, including: A gas sensor network and environmental parameter sensing units are deployed in and around the hot work area. The gas sensor network includes fixed gas monitoring nodes deployed at high-risk locations or on key equipment and mobile gas monitoring nodes mounted on a mobile platform. Based on the real-time data collected by the gas sensor network and environmental parameter sensing unit, a dynamic diffusion distribution field of combustible gas within the work area is constructed using computational fluid dynamics and Gaussian diffusion models. Based on the predicted values ​​of combustible gas concentration at each spatial location output by the dynamic diffusion distribution field of combustible gas and the prediction uncertainty characterizing the prediction confidence, the target area indicated by the concentration gradient direction of combustible gas or the prediction uncertainty is determined. The mobile gas monitoring node is controlled to move to the target area for patrol monitoring, and the verification concentration data is obtained. Based on the deviation between the verification concentration data and the predicted concentration at the same location by the dynamic diffusion distribution field of combustible gas, the computational fluid dynamics model is corrected, and the corrected dynamic diffusion distribution field of combustible gas is obtained. Centered on the real-time location of the hot work operation, and combined with the corrected dynamic diffusion distribution field of combustible gas, a reduced-order local response model is established to predict the concentration sequence of combustible gas around the hot work point. The solution period of the reduced-order local response model is shorter than that of the computational fluid dynamics model. Within a future preset time window, if any predicted value output by the reduced-order local response model is greater than or equal to the first safety threshold, a first linkage control command is generated; if any predicted value output by the reduced-order local response model is greater than or equal to the second safety threshold, or if the ignition point concentration value output by the corrected combustible gas dynamic diffusion distribution field is greater than or equal to the second safety threshold, a second linkage control command is generated; the second safety threshold is higher than the first safety threshold. The modified computational fluid dynamics model operates at a first frequency, the reduced-order local response model operates at a second frequency, the second frequency is higher than the first frequency, and the length of the future preset time window is greater than the update cycle corresponding to the first frequency.

[0022] In the above technical solution, the computational fluid dynamics (CFD) model is a mature numerical simulation tool in the field of fluid mechanics. Its basic theoretical system (Navier-Stokes equations, turbulence models, numerical discretization methods, etc.) has long been publicly available and belongs to common knowledge and existing technology. The Gaussian diffusion model (also known as the Gaussian plume model) is a classic mathematical model describing the diffusion law of atmospheric pollutants. Its core assumption is that the pollutant concentration follows a Gaussian (normal) distribution in both the horizontal and vertical directions.

[0023] In the above technical solution, the fixed gas monitoring node is used to provide continuous gas concentration time series for the corresponding location; the mobile gas monitoring node is used to conduct mobile detection of the monitoring blind zone of the fixed gas monitoring node or the area of ​​interest indicated by the model; the environmental parameter sensing unit is used to collect environmental wind speed and wind direction data in real time. The combustible gas dynamic diffusion distribution field outputs the predicted value of combustible gas concentration at each spatial location and the prediction uncertainty characterizing the prediction confidence level.

[0024] In the above technical solution, the deployment of the fixed gas monitoring nodes at the high-risk locations or key equipment meets the following conditions: the distance between two adjacent fixed gas monitoring nodes is not greater than the diameter of the detectable concentration range calculated by the Gaussian diffusion model under the preset minimum leakage rate and average wind speed conditions.

[0025] In the above technical solution, the prediction uncertainty characterizing the prediction confidence level is the variance or standard deviation of the concentration prediction values ​​output by each ensemble member at the same spatial location during the generation process of the combustible gas dynamic diffusion distribution field using the ensemble forecasting method. The prediction uncertainty is obtained through the following specific forecasting method: Based on the current leakage source parameters (location, velocity) and boundary conditions (wind speed, wind direction), random perturbations conforming to a normal distribution are applied to them respectively, generating N ensemble members (N ranges from 20 to 50). Each member independently runs a CFD model to obtain the predicted combustible gas concentration values ​​at each spatial location. The sample variance of the N predicted values ​​at the same spatial location is calculated, and this variance (or standard deviation) is used as the prediction uncertainty for that location. The perturbation amplitude is set as follows: the standard deviation of the leakage rate perturbation is 10% of the current estimate, the standard deviation of the leakage source location perturbation is 1m, the standard deviation of the wind speed perturbation is 0.2m / s, and the standard deviation of the wind direction perturbation is 5°. The first frequency is determined based on the minimum value among the sampling frequency of the fixed gas monitoring node, the sampling frequency of the environmental parameter sensing unit, and the computation time required for the computational fluid dynamics model to complete a single global solution. The second frequency is an integer multiple of the first frequency, and satisfies the condition that the single solution time of the reduced-order local response model is less than the solution cycle corresponding to the second frequency. The length of the future preset time window is 1.5 to 3 times the update cycle corresponding to the first frequency. Both the first and second safety thresholds are dynamic thresholds, adjusted in real time according to the operation type, operation intensity parameters, and environmental wind speed and direction data of the hot work. Under conditions of low environmental wind speed and unfavorable gas diffusion, the first and second safety thresholds are reduced to improve the safety margin. In a specific embodiment, the first safety threshold is set to 10% of the lower explosive limit (LEL) of combustible gas, and the second safety threshold is set to 25% of the LEL. Under unfavorable diffusion conditions with an environmental wind speed below 0.5 m / s, the safety thresholds are dynamically reduced: the first safety threshold is reduced to 8% LEL, and the second safety threshold is reduced to 20% LEL.

[0026] In the above technical solution, the step of correcting the computational fluid dynamics model based on the deviation between the verified concentration data and the predicted concentration at the same location of the dynamic diffusion distribution field of combustible gas specifically includes: using the deviation as the observation increment, and synchronously updating at least one of the leakage source parameters, boundary condition parameters, and diffusion coefficient parameters of the computational fluid dynamics model using an ensemble Kalman filter method. After correcting the computational fluid dynamics model and obtaining the corrected dynamic diffusion distribution field of combustible gas, the method further includes: performing a consistency check on the dynamic diffusion distribution field of combustible gas before and after correction; if the difference between the predicted concentration values ​​at the ignition point before and after correction exceeds a preset threshold, then triggering the reinitialization of the reduced-order local response model.

[0027] Specifically, the state vector is defined as x = [x0, y0, z0, Q, u, v, w, ε]ᵀ, where (x0, y0, z0) is the location of the leakage source, Q is the leakage rate, (u, v) are the wind speed components, w is the turbulent kinetic energy, and ε is the turbulent dissipation rate. The initial ensemble members are generated by the aforementioned ensemble prediction method. The observation vector y represents the concentration value measured by the moving node at the verification location. The observation matrix H maps the state to the predicted concentration values ​​at the observation locations. The update formula for the ensemble Kalman filter is: K is the Kalman gain matrix. ;P （f） The forecast error covariance matrix is ​​obtained by combining membership statistics; R is the observation error covariance. The updated state vector is used to rerun the CFD model to obtain the corrected diffusion distribution field.

[0028] In practice, the process also includes a sensor node self-diagnosis step: real-time monitoring of the operating status parameters of each fixed and mobile gas monitoring node. When any node's operating status parameter becomes abnormal, the node is marked as a faulty node, and its data is removed from the node set participating in the construction of the combustible gas dynamic diffusion distribution field. Once a fixed gas monitoring node is marked as a faulty node, if necessary, the data fusion weight of effective fixed gas monitoring nodes surrounding the faulty node can be increased, and a scheduling command can be generated to prioritize controlling the mobile gas monitoring nodes to travel to the location of the faulty node for replacement monitoring.

[0029] In the above technical solution, before constructing the dynamic diffusion distribution field of combustible gas based on the data collected in real time by the gas sensor network and the environmental parameter sensing unit, the real-time collected data is preprocessed. The preprocessing includes: removing outliers that exceed the sensor range, interpolating and filling missing data, and smoothing and denoising the data using moving average filtering or Kalman filtering.

[0030] The above technical solution establishes a hierarchical architecture of global slow update of CFD model and local fast prediction of reduced-order local response model. It uses mobile node survey data to perform closed-loop verification and correction of diffusion distribution field, and triggers graded safety interlocks in advance based on the predicted future concentration of ignition point. This solves the core contradiction in the existing single-point threshold triggering method where safety response lags behind gas diffusion speed and frequent malfunctions are difficult to balance, and achieves a technical leap from passive response to active prediction.

[0031] In one technical solution, the first linkage control command execution includes an audible and visual warning action, and the second linkage control command execution includes a power-off and flameout action and a gas supply cut-off action. After generating the second linkage control command and executing the power-off and flameout and gas supply cut-off emergency cut-off actions, the process further includes: continuously monitoring the corrected dynamic diffusion distribution field of combustible gas; when the predicted value of combustible gas concentration at the ignition point is continuously lower than the first safety threshold and reaches a preset safety duration, a reset command is generated, allowing the hot work to resume. The linkage control command is divided into two levels: the first level triggers an audible and visual warning to remind operators to pay attention to the risk situation, and the second level triggers the power-off and flameout and gas supply cut-off emergency cut-off actions. This achieves graded handling of warnings and cut-offs, avoiding unnecessary interruptions to production continuity caused by direct cut-offs, and balancing safety protection and operational efficiency.

[0032] In one technical solution, the data collected in real time by the gas sensor network and environmental parameter sensing unit includes the node locations of the gas sensor network, real-time gas concentration data from the fixed gas monitoring nodes, wandering detection concentration data from the mobile gas monitoring nodes, and environmental wind speed and direction data. By uniformly incorporating the continuous monitoring data from fixed nodes, the wandering detection data from mobile nodes, and the environmental wind speed and direction data into the construction input of the diffusion distribution field, spatiotemporal fusion of multi-source heterogeneous sensing data is achieved. This allows the dynamic diffusion distribution field of combustible gases to reflect the temporal evolution of fixed locations and to compensate for the spatial coverage deficiencies of fixed nodes through mobile blind spot detection.

[0033] In one of the technical solutions, the construction of the dynamic diffusion distribution field of combustible gas within the operating area specifically includes: The data inputs include the node locations of the gas sensor network, the real-time gas concentration data of the fixed gas monitoring nodes, the mobile gas monitoring node's floating detection concentration data, and the ambient wind speed and direction data. The environmental wind speed and direction data and the real-time gas concentration data collected by the fixed gas monitoring node are used as inputs to the Gaussian diffusion model. The location and leakage rate of the leakage source are used as unknown variables to solve the Gaussian diffusion model in reverse, and the preliminary location and leakage rate of the leakage source are derived as leakage source parameters. Using the leakage source parameters and the ambient wind speed and direction data as inputs to the Gaussian diffusion model, the initial estimated values ​​of combustible gas concentration at each spatial location within the work area are calculated to form an initial diffusion distribution field. Using the initial diffusion distribution field as the initial condition of the computational fluid dynamics model, and combining the three-dimensional geometric model and boundary conditions of the work area, the computational fluid dynamics model is used for numerical solution to obtain the dynamic diffusion distribution field of combustible gas. The dynamic diffusion distribution field of combustible gas outputs the predicted values ​​of combustible gas concentration at each spatial location and the prediction uncertainty characterizing the prediction confidence level. The three-dimensional geometric model of the work area is obtained through one of the following methods: 1. Scanning the site with a three-dimensional laser scanner to generate point cloud data, and then forming a computational grid through meshing; 2. Based on the CAD design drawings of the work site, extracting the geometric dimensions of equipment, pipelines, and structures, and using preprocessing software (such as Gambit, Pointwise) to construct an unstructured mesh. Boundary conditions include: inlet boundary (wind speed, wind direction, turbulence intensity), outlet boundary (pressure outlet), wall boundary (no slip or slip condition), and leakage source boundary (given leakage rate and composition). Ambient wind speed and direction data are provided in real time by an ultrasonic anemometer as time-varying parameters of the inlet boundary.

[0034] In the above technical solution, before constructing the dynamic diffusion distribution field of combustible gas, the method further includes: based on the gas concentration time series collected by the fixed gas monitoring node, determining whether there is a situation where the combustible gas concentration exceeds a preset background threshold or the concentration rise rate exceeds a preset rate threshold; if it is determined that there is no such situation, it is determined that there is no leakage, the normal monitoring state is maintained, and the subsequent diffusion distribution field construction and linkage control steps are not initiated; if it is determined that there is such situation, it is determined that there is a leakage, and the subsequent construction of the dynamic diffusion distribution field of combustible gas and subsequent steps are initiated.

[0035] In the above technical solution, during the numerical solution of the computational fluid dynamics model, the real-time gas concentration data of the fixed gas monitoring node is used as an internal observation constraint, and the solution process of the computational fluid dynamics model is corrected in real time through a data assimilation algorithm.

[0036] In the above technical solution, the inverse solution of the Gaussian diffusion model to deduce the preliminary location and leakage rate of the leak source is specifically achieved using one of the following methods: an optimization method based on gradient descent or genetic algorithm, with the objective function being minimizing the error between the real-time gas concentration data collected by the fixed gas monitoring node and the concentration calculated by the Gaussian diffusion model in the forward direction, iteratively searching for the location and leakage rate of the leak source; or a source tracing method based on the adjoint equation, starting from the location of the fixed gas monitoring node where the gas concentration anomaly is detected, solving the adjoint equation of the Gaussian diffusion model in the reverse direction along the environmental wind speed and direction data, and deducing the location and leakage rate of the leak source. The data assimilation algorithm is an ensemble Kalman filter method or a three-dimensional variational assimilation method; through the data assimilation algorithm, the real-time gas concentration data of the fixed gas monitoring node is used as the observation value to update the state variables or parameters in the computational fluid dynamics model solution process in real time, so that the dynamic diffusion distribution field of the combustible gas approximates the measured concentration.

[0037] Taking gradient descent as an example, the objective function is defined as follows: Where P = (x0, y0, z0, Q) T Let Q be the location and rate of the leak source, and Q be the leak rate; K be the number of fixed nodes that detected abnormal concentrations, and C be the number of fixed nodes that detected abnormal concentrations. i Let C be the measured concentration at the i-th node. Gauss (r) i ;P) represents the Gaussian diffusion model at position r given p. i The positive concentration value is calculated at p. The gradient of the objective function with respect to p is calculated using the finite difference method. p is updated along the negative gradient direction, with the step size following the Armijo criterion. Iteration continues until the gradient magnitude is less than 1 × 10⁻⁶. -4 Or it can reach the maximum number of iterations of 100. If a genetic algorithm is used, the population size is set to 50, the crossover probability is 0.8, the mutation probability is 0.1, and the number of generations is 100.

[0038] In the above technical solution, when the initial diffusion distribution field is used as the initial condition of the computational fluid dynamics model, the relaxation iteration method is used to gradually embed the initial diffusion distribution field into the solution mesh of the computational fluid dynamics model, so as to avoid numerical oscillations caused by incompatibility between the initial field and the boundary conditions of the CFD model.

[0039] The above technical solution adopts a progressive construction strategy: first, the Gaussian diffusion model is used to solve for the leakage source parameters in reverse; then, the initial diffusion distribution field is calculated in forward; and finally, a CFD model is used for high-precision numerical solution. The Gaussian model with low computational cost is used to quickly locate the approximate location and intensity of the leakage source. Then, this model is used to drive the CFD model with high computational cost for detailed simulation. This significantly reduces the computational overhead and convergence difficulty of the CFD model cold start while ensuring the accuracy of the diffusion distribution field.

[0040] In one of the technical solutions, the determination of the target area indicated by the concentration gradient direction of the combustible gas or the prediction uncertainty specifically includes: The predicted values ​​of combustible gas concentration at each spatial location are extracted from the dynamic diffusion distribution field of the combustible gas. The partial derivatives of the predicted values ​​of combustible gas concentration in each spatial direction are calculated to obtain the concentration gradient vector field of the combustible gas. The direction of each vector in the concentration gradient vector field points to the direction in which the combustible gas concentration rises the fastest at the corresponding location. The prediction uncertainty of each spatial location is extracted from the dynamic diffusion distribution field of the combustible gas, and the region where the prediction uncertainty exceeds a preset uncertainty threshold is identified as a low confidence region of the dynamic diffusion distribution field of the combustible gas. Based on the concentration gradient vector field and the low confidence region, the target region is determined according to priority: the region along the direction indicated by the concentration gradient vector field is taken as the first priority target region, and the part of the low confidence region of the model that does not overlap with the first priority target region is taken as the second priority target region; The mobile gas monitoring node is controlled to move to the first priority target area for patrol monitoring first, and after completing the patrol monitoring in the first priority target area, it moves to the second priority target area for patrol monitoring.

[0041] In the above technical solution, the method for determining the preset uncertainty threshold is as follows: statistically analyze the distribution of predicted uncertainties at all spatial locations in the dynamic diffusion distribution field of the combustible gas, and set the preset uncertainty threshold as a preset multiple of the mean of the predicted uncertainties, or as a preset quantile of the distribution. Furthermore, the preset uncertainty threshold is dynamically adjusted according to the type and intensity parameters of the current hot work operation; wherein, the higher the risk level of the hot work operation, the lower the preset uncertainty threshold is set.

[0042] In the above technical solution, the area along the location indication direction of the concentration gradient vector is designated as the first priority target area. Specifically, this includes taking the node position in the fixed gas monitoring node where the gas concentration exceeds a preset background threshold as the starting point, extending a preset distance along the direction of the concentration gradient vector at that position, and defining the area within the extended path and its surrounding preset radius as the first priority target area.

[0043] In the above technical solution, controlling the mobile gas monitoring node to move to the target area for patrol monitoring specifically includes: Obtain the position coordinates of the first priority target region and the second priority target region; Combining obstacle information and environmental wind speed and direction data within the operating area, a feasible path is first planned for the mobile gas monitoring node from its current location to the first priority target area. The mobile gas monitoring node is then controlled to move along this path. Upon reaching the first priority target area, it performs a spiral or grid-shaped scanning concentration detection within the first priority target area to obtain verification concentration data. The scanning interval of the spiral or grid-shaped trajectory is dynamically determined based on the response time of the gas sensor mounted on the mobile gas monitoring node and the moving speed of the mobile platform, ensuring that the dwell time at each scanning point is not less than the response time of the gas sensor. Let the response time of the mounted gas sensor be τ (unit: seconds), and the moving speed of the mobile platform be v (unit: m / s). Then the scanning interval Δd should satisfy: Δd ≤ v·τ, to ensure that the sensor has sufficient dwell time at each sampling point to achieve a stable reading. For example, if the sensor response time τ = 2s and the moving speed v = 0.2m / s, then Δd ≤ 0.4m, and in practice, Δd = 0.3m is taken. The pitch of the spiral trajectory is set to Δd, and the radius of the coverage area is determined according to the size of the target area; the grid spacing of the grid trajectory is also set to Δd. When encountering obstacles, the Dynamic Window Method (DWA) is used for local obstacle avoidance. After completing the survey of the first priority target area, a feasible path is planned for the mobile gas monitoring node to move from the first priority target area to the second priority target area. The mobile gas monitoring node is then controlled to move to the second priority target area and perform scanning concentration detection in the area using a spiral or grid-shaped trajectory to obtain the verification concentration data in the second priority target area.

[0044] In the above technical solution, when there are multiple fixed gas monitoring nodes that detect gas concentrations exceeding a preset background threshold, the node with the highest concentration value is selected as the starting point; or, candidate regions are defined with each node that detects gas concentrations exceeding the preset background threshold as the starting point, and the union of the candidate regions is taken as the first priority target region.

[0045] In the above technical solution, by combining the obstacle information in the working area and the environmental wind speed and direction data, a feasible path from the mobile gas monitoring node to the target area is planned. Specifically, the A* path planning algorithm or the fast expanding random tree algorithm is adopted, with obstacle avoidance and shortest path as the optimization objectives. The path is also modified by combining the environmental wind speed and direction data, so that the mobile gas monitoring node approaches the target area from the downwind or crosswind direction of the leak source.

[0046] In the above technical solution, when there are multiple mobile gas monitoring nodes, they are coordinated and scheduled: the first priority target area is assigned to the mobile gas monitoring node that is closest or has the strongest endurance, and the second priority target area is assigned to the remaining mobile gas monitoring nodes to achieve parallel monitoring. During the monitoring process, if the gas sensor onboard a mobile gas monitoring node detects that the concentration of combustible gas exceeds a preset emergency threshold, it immediately terminates the current monitoring task and transmits its current location coordinates and concentration data back in real time to trigger an emergency update of the dynamic diffusion distribution field of the combustible gas.

[0047] The above technical solution calculates the concentration gradient vector field and combines it with the prediction uncertainty. It uses a dual-criteria priority mechanism to determine the target area of ​​the mobile node's patrol. The gradient direction indicates the location of potential leakage sources, and the high uncertainty area indicates the places where the model needs to be filled and verified. This realizes the autonomous optimization and scheduling of the patrol path of the mobile gas monitoring node, so that each mobile detection can maximize its information gain for improving the overall accuracy of the diffusion distribution field.

[0048] In one of the technical solutions, the establishment of the reduced-order local response model specifically includes: Centered on the real-time location of the hot work operation, local concentration distribution data within a preset spatial range around the real-time location is extracted from the corrected dynamic diffusion distribution field of combustible gas and used as a local concentration snapshot sample. For the snapshot matrix composed of local concentration snapshot samples at multiple time points, an intrinsic orthogonal decomposition method is used to extract a set of dominant spatial modes, which characterize the main spatial features of the gas concentration distribution around the ignition point; Based on the dominant spatial modes, the computational fluid dynamics control equations are subjected to Galerkin projection to obtain a set of low-dimensional ordinary differential equations about the modal coefficients. The global diffusion trend output by the computational fluid dynamics model after real-time parameter assimilation and correction, when running at the first frequency, is used as the time-varying boundary condition of the low-dimensional ordinary differential equation system. The low-dimensional ordinary differential equation system is numerically integrated at the second frequency to obtain the time series of modal coefficients. The time series of modal coefficients is superimposed with the dominant spatial mode to reconstruct the gas concentration prediction sequence of the ignition point within the future preset time window.

[0049] In the above technical solution, the determination of the preset spatial range includes: the operation type and intensity parameters of the hot work, and the concentration gradient distribution around the hot work point in the corrected dynamic diffusion distribution field of combustible gas; wherein, the greater the intensity of the hot work and the higher the risk of the operation type, the larger the preset spatial range is set. Specifically, the preset radius of the preset spatial range is determined as follows: taking the real-time location of the hot work as the starting point, extending outward along the gradient direction at that location in the concentration gradient vector field until the predicted value of the combustible gas concentration decays to a position below the preset safe concentration threshold, and using this extension distance as the preset radius; if there is no significant concentration gradient direction, a preset default radius is used; or, the preset radius is not less than the maximum possible transport distance of the environmental wind speed data within an update cycle corresponding to the first frequency.

[0050] In the above technical solution, the sampling time of the local concentration snapshot sample is synchronized with the update time of the computational fluid dynamics model after real-time parameter assimilation correction running at the first frequency, so as to ensure that each snapshot sample is based on the latest corrected global diffusion trend.

[0051] In the above technical solution, the snapshot matrix is ​​composed of local concentration snapshot samples arranged in time sequence. Before performing intrinsic orthogonal decomposition, the time mean of each column of the snapshot matrix is ​​subtracted to extract the dominant spatial mode of the pulsation. The number of dominant spatial modes is determined according to the modal energy ratio: the modes are arranged in descending order of singular values, and the top few modes whose cumulative energy ratio exceeds a preset percentage threshold are selected as the dominant spatial modes. When the global diffusion trend output by the computational fluid dynamics model after real-time parameter assimilation correction changes significantly, causing the prediction deviation of the reduced-order local response model to exceed the preset tolerance, the local concentration snapshot sample acquisition and intrinsic orthogonal decomposition are re-executed to reconstruct the reduced-order local response model online. In specific implementation, a spherical region with a radius of 5m centered on the ignition point is used as the preset spatial range, and the concentration values ​​on the grid nodes within this region are used to form a spatial vector. Local concentration snapshots are collected for the M CFD update times preceding the current time (M ranges from 50 to 100). Each snapshot is arranged column-wise to form a snapshot matrix U ∈ R^{N×M}, where N is the number of local spatial grid points. Singular value decomposition is performed on U, and the first r left singular vectors are taken as the dominant spatial modes, where r is determined by the cumulative energy percentage. The CFD governing equations are projected onto these r modes to obtain a set of low-dimensional ordinary differential equations for the mode coefficients a(t) ∈ R^r: da / dt = F(a, t), where F is the nonlinear term after projection. The fourth-order Runge-Kutta method is used for numerical integration at the second frequency to obtain the time series of mode coefficients, and then the concentration prediction value series is reconstructed.

[0052] The above technical solution uses the intrinsic orthogonal decomposition method to reduce the order of the full-order CFD model. By extracting the dominant spatial modes and Galerkin projection, a low-dimensional set of ordinary differential equations is constructed. While preserving the accuracy of the CFD model in characterizing the physical process of gas diffusion, the solution cycle is compressed to the second level or even the sub-second level, thus solving the contradiction between the computation time of high-precision CFD models and the need for rapid linkage response at the millisecond level.

[0053] In one technical solution, the alternative to the intrinsic orthogonal decomposition method is the dynamic mode decomposition method. When using the dynamic mode decomposition method, dynamic modes describing the spatiotemporal evolution of gas concentration around the ignition point and their corresponding eigenvalues ​​are extracted from the snapshot matrix. A low-dimensional linear evolution model is constructed based on the dynamic modes and eigenvalues ​​to replace the Galerkin projection step, directly extrapolating and predicting the gas concentration prediction sequence within the future preset time window. This application provides the dynamic mode decomposition method as an alternative to intrinsic orthogonal decomposition. The DMD method does not require Galerkin projection and can directly extract the spatiotemporally evolving dynamic modes and eigenvalues ​​from the snapshot matrix and extrapolate for prediction. This results in a simpler implementation path and higher computational efficiency, providing an optional lightweight technical route for constructing reduced-order local response models and enhancing the adaptability of the solution under different computing platforms and application scenarios.

[0054] In one of the technical solutions, the environmental parameter sensing unit includes at least one ultrasonic anemometer; the fixed gas monitoring node and the mobile gas monitoring node employ at least one of the following sensor types: The catalytic combustion combustible gas sensor is used to detect hydrocarbon combustible gases. Its output signal has a near-linear relationship with the concentration of combustible gas and has a broad spectrum response characteristic to various combustible gases. Infrared absorption combustible gas sensors are used to detect specific combustible gases. They are based on the principle that infrared light of a specific wavelength is absorbed by the target gas molecules and are not affected by the susceptibility of catalytic combustion sensors to poisoning by sulfides or halogen compounds. Electrochemical gas sensors are used to detect the concentration of specific toxic gases or oxygen. Semiconductor-based combustible gas sensors utilize the principle of changes in the conductivity of metal oxide semiconductor materials in a combustible gas environment for detection. The fixed gas monitoring node is equipped with an explosion-proof enclosure with an explosion-proof rating of not less than ExdIICT6 and a protection rating of not less than IP65. The sampling method is either diffusion or pump suction.

[0055] In the above technical solution, the deployment locations of the fixed gas monitoring nodes include: Flange connections of critical equipment, valve sealing points, pump body sealing points, and breather valve outlets of storage tanks around the hot work site; Locations of pipe welds, joints, and potential leak sources within a predetermined radius centered on the hot work site; Low-lying areas or ventilation dead zones in the work area where gas tends to accumulate.

[0056] In the above technical solution, the fixed gas monitoring nodes and the mobile gas monitoring nodes in the gas sensor network adopt a unified wireless communication protocol and form a mesh network through a wireless self-organizing network to realize data interaction and collaborative scheduling between nodes and between nodes and the system platform.

[0057] In the above technical solution, the mobile platform is one of the following types: The explosion-proof wheeled inspection robot uses an electric four-wheel drive chassis and is equipped with 3D laser SLAM technology to achieve autonomous navigation and obstacle avoidance. Explosion-proof tracked inspection robot for use in complex terrain environments; The explosion-proof quadrupedal bionic robot (robot dog) adopts a quadrupedal bionic structure and can overcome obstacles such as pipes, steps, and steep slopes. Explosion-proof drones are multi-rotor or vertical take-off and landing drones equipped with gas detection modules and thermal imagers, enabling them to perform aerial detection, leak source location and data transmission tasks over hazardous areas. The mobile platform is explosion-proof certified, with an explosion-proof rating of no less than ExdIIBT4Gb.

[0058] The above technical solution, by limiting the environmental sensing configuration of ultrasonic anemometers and allowing for the optional combination of various types of gas sensors such as catalytic combustion, infrared absorption, electrochemical, and semiconductor sensors, and by specifying technical requirements such as the explosion-proof level of fixed nodes being no less than ExdIICT6 and the protection level being no less than IP65, ensures the long-term reliable operation of the sensor network in harsh industrial environments such as petrochemical plants, which contain deflagration hazards and corrosive atmospheres, from the hardware level. This provides a stable and accurate data foundation for the construction of the upper diffusion field and the linkage control algorithm.

[0059] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A gas sensor network and hot work linkage control method, characterized in that, include: A gas sensor network and environmental parameter sensing units are deployed in and around the hot work area. The gas sensor network includes fixed gas monitoring nodes deployed at high-risk locations or on key equipment and mobile gas monitoring nodes mounted on mobile platforms. Based on the data collected in real time by the gas sensor network and environmental parameter sensing unit, a dynamic diffusion distribution field of combustible gas in the working area is constructed based on the computational fluid dynamics model and the Gaussian diffusion model. Based on the predicted values ​​of combustible gas concentration at each spatial location output by the dynamic diffusion distribution field of combustible gas and the prediction uncertainty characterizing the prediction confidence, the target area indicated by the concentration gradient direction of combustible gas or the prediction uncertainty is determined. The mobile gas monitoring node is controlled to move to the target area for patrol monitoring, and the verification concentration data is obtained. Based on the deviation between the verification concentration data and the predicted concentration at the same location by the dynamic diffusion distribution field of combustible gas, the computational fluid dynamics model is corrected, and the corrected dynamic diffusion distribution field of combustible gas is obtained. Centered on the real-time location of the hot work operation, and combined with the corrected dynamic diffusion distribution field of combustible gas, a reduced-order local response model is established to predict the concentration sequence of combustible gas around the hot work point. The solution period of the reduced-order local response model is shorter than that of the computational fluid dynamics model. Within a future preset time window, if any predicted value output by the reduced-order local response model is greater than or equal to the first safety threshold, a first linkage control command is generated; if any predicted value output by the reduced-order local response model is greater than or equal to the second safety threshold, or if the ignition point concentration value output by the corrected combustible gas dynamic diffusion distribution field is greater than or equal to the second safety threshold, a second linkage control command is generated; the second safety threshold is higher than the first safety threshold. The modified computational fluid dynamics model runs at a first frequency, the reduced-order local response model runs at a second frequency, the second frequency is higher than the first frequency, and the length of the future preset time window is greater than the update cycle corresponding to the first frequency.

2. The gas sensor network and hot work linkage control method as described in claim 1, characterized in that, The first linkage control command execution includes an audible and visual warning action, and the second linkage control command execution includes a power-off and flameout action and a gas supply cut-off action.

3. The gas sensor network and hot work linkage control method as described in claim 2, characterized in that, The data collected in real time by the gas sensor network and environmental parameter sensing unit includes the node locations of the gas sensor network, the real-time gas concentration data of the fixed gas monitoring nodes, the mobile gas monitoring node's floating detection concentration data, and the environmental wind speed and direction data.

4. The gas sensor network and hot work linkage control method as described in claim 3, characterized in that, The construction of the dynamic diffusion distribution field of combustible gas within the operating area specifically includes: The data inputs include the node locations of the gas sensor network, the real-time gas concentration data of the fixed gas monitoring nodes, the mobile gas monitoring node's floating detection concentration data, and the ambient wind speed and direction data. The environmental wind speed and direction data and the real-time gas concentration data collected by the fixed gas monitoring node are used as inputs to the Gaussian diffusion model. The location and leakage rate of the leakage source are used as unknown variables to solve the Gaussian diffusion model in reverse, and the preliminary location and leakage rate of the leakage source are derived as leakage source parameters. Using the leakage source parameters and the ambient wind speed and direction data as inputs to the Gaussian diffusion model, the initial estimated values ​​of combustible gas concentration at each spatial location within the work area are calculated to form an initial diffusion distribution field. Using the initial diffusion distribution field as the initial condition of the computational fluid dynamics model, and combining the three-dimensional geometric model and boundary conditions of the working area, the computational fluid dynamics model is used to perform numerical solutions to obtain the dynamic diffusion distribution field of the combustible gas; the dynamic diffusion distribution field of the combustible gas outputs the predicted values ​​of the combustible gas concentration at each spatial location and the prediction uncertainty characterizing the prediction confidence.

5. The gas sensor network and hot work linkage control method as described in claim 4, characterized in that, The determination of the target region indicated by the direction of the concentration gradient of combustible gas or the prediction uncertainty specifically includes: The predicted values ​​of combustible gas concentration at each spatial location are extracted from the dynamic diffusion distribution field of the combustible gas. The partial derivatives of the predicted values ​​of combustible gas concentration in each spatial direction are calculated to obtain the concentration gradient vector field of the combustible gas. The direction of each vector in the concentration gradient vector field points to the direction in which the combustible gas concentration rises the fastest at the corresponding location. The prediction uncertainty of each spatial location is extracted from the dynamic diffusion distribution field of the combustible gas, and the region where the prediction uncertainty exceeds a preset uncertainty threshold is identified as a low confidence region of the dynamic diffusion distribution field of the combustible gas. Based on the concentration gradient vector field and the low confidence region, the target region is determined according to priority: the region along the direction indicated by the concentration gradient vector field is taken as the first priority target region, and the part of the low confidence region of the model that does not overlap with the first priority target region is taken as the second priority target region; The mobile gas monitoring node is controlled to move to the first priority target area for patrol monitoring first, and after completing the patrol monitoring in the first priority target area, it moves to the second priority target area for patrol monitoring.

6. The gas sensor network and hot work linkage control method as described in claim 5, characterized in that, The establishment of the reduced-order local response model specifically includes: Centered on the real-time location of the hot work operation, local concentration distribution data within a preset spatial range around the real-time location is extracted from the corrected dynamic diffusion distribution field of combustible gas and used as a local concentration snapshot sample. For the snapshot matrix composed of local concentration snapshot samples at multiple time points, an intrinsic orthogonal decomposition method is used to extract a set of dominant spatial modes, which characterize the main spatial features of the gas concentration distribution around the ignition point; Based on the dominant spatial modes, the computational fluid dynamics control equations are subjected to Galerkin projection to obtain a set of low-dimensional ordinary differential equations about the modal coefficients. The global diffusion trend output by the computational fluid dynamics model after real-time parameter assimilation and correction, when running at the first frequency, is used as the time-varying boundary condition of the low-dimensional ordinary differential equation system. The low-dimensional ordinary differential equation system is numerically integrated at the second frequency to obtain the time series of modal coefficients. The time series of modal coefficients is superimposed with the dominant spatial mode to reconstruct the gas concentration prediction sequence of the ignition point within the future preset time window.

7. The gas sensor network and hot work linkage control method as described in claim 6, characterized in that, The alternative to the intrinsic orthogonal decomposition method is the dynamic mode decomposition method. When using the dynamic mode decomposition method, dynamic modes describing the spatiotemporal evolution of gas concentration around the ignition point and corresponding eigenvalues ​​are extracted from the snapshot matrix. A low-dimensional linear evolution model is constructed based on the dynamic modes and eigenvalues ​​to replace the Galerkin projection step and directly extrapolate to predict the gas concentration prediction sequence within the future preset time window.

8. The gas sensor network and hot work linkage control method as described in claim 1, characterized in that, The environmental parameter sensing unit includes at least one ultrasonic anemometer; the fixed gas monitoring node and the mobile gas monitoring node employ at least one of the following sensor types: The catalytic combustion combustible gas sensor is used to detect hydrocarbon combustible gases. Its output signal has a near-linear relationship with the concentration of combustible gas and has a broad spectrum response characteristic to various combustible gases. Infrared absorption combustible gas sensors are used to detect specific combustible gases. They are based on the principle that infrared light of a specific wavelength is absorbed by the target gas molecules and are not affected by the susceptibility of catalytic combustion sensors to poisoning by sulfides or halogen compounds. Electrochemical gas sensors are used to detect the concentration of specific toxic gases or oxygen. Semiconductor-based combustible gas sensors utilize the principle of changes in the conductivity of metal oxide semiconductor materials in a combustible gas environment for detection. The fixed gas monitoring node is equipped with an explosion-proof enclosure with an explosion-proof rating of not less than ExdIICT6 and a protection rating of not less than IP65. The sampling method is either diffusion or pump suction.

Images