Metallurgical gas leakage tracing method and system based on physical inversion and optimization solution

Abstract

A metallurgical gas leakage source tracing method and system based on physical inversion and optimization solution, the method comprising: acquiring plant geographic information, meteorological data and sensor layout information, completing data preprocessing and time sequence alignment, and constructing a plant three-dimensional diffusion physical model for leakage source tracing; taking leakage source parameters as decision variables, and minimizing the deviation between sensor observation concentration and diffusion model theoretical concentration as the target, a constrained source inversion optimization problem is established; a two-stage progressive optimization strategy of coarse search first and fine search later is adopted; the optimal leakage source parameters are output, and the source tracing results are continuously verified by using new observation data; based on the leakage source position and leakage parameters obtained by source tracing, emergency response decisions are automatically triggered. The system comprises a plant physical model construction module, a source inversion optimization problem modeling module, a hierarchical optimization solution module, an online verification and adaptive correction module, and a source-driven emergency linkage module to form a source tracing system. The present application can realize stable and reliable leakage source positioning.

Description

Technical Field

[0001] This invention belongs to the field of metallurgical safety production monitoring and intelligent early warning technology, specifically involving a method and system for tracing the source of metallurgical gas leaks based on physical inversion and optimization solution. Background Technology

[0002] Metal smelting is a pillar industry of the national economy, but it is also a typical high-risk production industry. Core metallurgical equipment such as blast furnaces, converters, gas pipelines, and gas holders operate under complex conditions of high temperature, high pressure, and strong corrosion, making them highly susceptible to leaks of toxic and harmful gases such as carbon monoxide and hydrogen sulfide. These accidents are characterized by their suddenness, rapid spread, wide impact, and high hazard. In recent years, gas leak-related poisoning and explosion accidents have accounted for a consistently high proportion of major and above-level safety accidents in the metallurgical industry, making them a core risk source restricting safe production in metallurgical plants. Therefore, achieving accurate source tracing, early prediction, and dynamic hazard simulation of toxic and harmful gas leaks in metallurgical plants is a key technological requirement for improving plant emergency response capabilities, avoiding major safety accidents, and ensuring the safety of personnel and equipment.

[0003] Currently, existing technologies for preventing and tracing leaks in the metallurgical industry have significant shortcomings, making them unsuitable for the tracing and prevention needs of complex industrial sites. Traditional monitoring methods often employ single-point sensor threshold alarm mechanisms, which are reactive and reactive. Due to the dense equipment, intricate pipelines, and complex layout of metallurgical plants, there are numerous monitoring blind spots. Furthermore, sensors are susceptible to interference from the complex environment, frequently resulting in false alarms and missed alarms. They can only output single-point concentration data, failing to provide effective data support for leak tracing, risk assessment, and emergency decision-making.

[0004] Existing technologies have significant limitations in leakage prediction and source tracing modeling. Traditional physical mechanism models are mostly based on ideal assumptions such as uniform wind fields and flat terrain, which cannot adapt to the actual working conditions of metallurgical plants with dense structures, unsteady wind fields, and turbulent flow fields. Shallow machine learning models are difficult to fit the nonlinear leakage patterns coupled by multiple factors such as temperature, wind pressure, equipment aging, and environmental interference. General deep learning models have not been specifically optimized for metallurgical leakage scenarios, and cannot effectively integrate multi-dimensional data such as plant geospatial information, equipment health status, and environmental monitoring. They are also unable to distinguish the differences in leakage characteristics between different equipment and different locations, resulting in insufficient model generalization ability and source tracing accuracy.

[0005] In summary, existing metallurgical gas leak prevention and control technologies have not formed an integrated source tracing and prevention system that integrates multi-source data fusion and multi-model collaboration. This limits their ability to overcome the technical limitations of traditional post-incident handling and hinders the proactive prediction, accurate source tracing, and dynamic hazard simulation of gas leaks in complex metallurgical scenarios. Addressing the shortcomings of existing technologies, such as low source tracing accuracy, poor real-time performance, weak adaptability to operating conditions, and insufficient intelligence, there is an urgent need for a metallurgical gas leak source tracing method based on physical inversion and optimization solutions. This method aims to resolve the pain points of existing technologies and improve the accuracy of gas leak source tracing and the level of safe production prevention and control in metallurgical plants. Summary of the Invention

[0006] To address the problems of existing technologies, this invention provides a method and system for tracing metallurgical gas leaks based on physical inversion and optimization. This method transforms the tracing of metallurgical gas leaks from experience-based judgment relying on historical data to quantitative inversion driven by real-time observation and physical models, achieving accurate leak source location and emergency decision-making closed loop under zero-sample conditions. The system reconstructs the actual diffusion process in the plant area through multi-field coupled physical modeling, achieves stable and reliable leak source location through hierarchical optimization and uncertainty quantification, and directly maps the tracing results to hierarchical emergency operation plans.

[0007] To achieve the above objectives, this invention provides a method for tracing the source of metallurgical gas leaks based on physical inversion and optimization solutions, comprising the following steps: Step 1: Construct a three-dimensional diffusion physical model of the plant area; acquire geographical information, meteorological data and sensor layout information of the plant area, complete data preprocessing and time series alignment, and construct a three-dimensional diffusion physical model of the plant area for leak tracing. Step 2: Construct a source tracing and inversion optimization problem; using the leakage source parameters as decision variables and minimizing the deviation between the concentration observed by the sensor and the theoretical concentration of the diffusion model as the objective, establish a constrained source tracing and inversion optimization problem; Step 3: Hierarchical optimization solution strategy; adopt a two-level progressive optimization strategy of first coarse search and then fine search, and add a global maximum computation time constraint to balance the timeliness of emergency response and the accuracy of source tracing; Step 4: Output and online verification of leakage source parameters and adaptive correction; output the optimal leakage source parameters and continuously verify the source tracing results using subsequent new observation data; Step 5: Emergency response linkage driven by source tracing results; Based on the location and parameters of the leak source obtained from the source tracing, an emergency response decision centered on the location of the leak source is automatically triggered.

[0008] As a preferred option, the process of constructing the three-dimensional diffusion physical model of the plant area in step 1 is as follows: S11: Obtain geographic information data of the plant area through the GIS system; call the system's built-in metallurgical standard component library to generate a three-dimensional geometric model of the plant area, and mark the leak-proof areas and potential leak equipment pipe sections in the plant area; S12: Collect the three-dimensional coordinates of all gas sensor installation locations, sensor models, measurement accuracy, and saturation threshold parameters; mark the sensor locations in the three-dimensional geometric model to form a sensor observation network; synchronously acquire real-time meteorological datasets, including wind speed, wind direction, temperature, humidity, and atmospheric stability level, and unify the timestamps of all monitoring data to complete time-series registration; S13: Construct a three-dimensional wind field in the plant area; using real-time meteorological data as boundary conditions, solve for the three-dimensional spatial wind field distribution in the plant area using a fast wind field diagnostic model or a pre-trained surrogate wind field model. The effects of building obstruction and airflow around the building are taken into account; a building density discrimination criterion is set to realize the automatic selection of wind field model: logarithmic wind profile is used for open areas, and a fast wind field parameterization scheme based on the simplified solution of RANS equation is used for densely built areas. S14: Construct a diffusion physical model; automatically select a forward diffusion model based on wind field stability quantification indicators; the wind field stability quantification indicators Wind direction stability component and wind speed stability component The weighted fusion yields: ,in , , and These represent the standard deviations of wind direction and wind speed within the preset time window. and For the preset threshold, For weighting coefficients; when When the wind field is considered stable, the Gaussian diffusion model is selected; when... When the wind field is determined to be non-stationary, the Lagrange particle diffusion model is selected; the diffusion model uses the leakage source parameter set. As input, output the theoretical concentration value at each sensor location. ,in The three-dimensional coordinates of the leakage source For leakage rate, The leak start time, The type of substance leaked.

[0009] As a preferred option, the process of constructing the source tracing and inversion optimization problem in step 2 is as follows: S21: Establish the observation-model bias function; for those with A network of gas sensors is used to define a weighted deviation function between the observed concentration and the theoretical concentration. : ; In the formula, For the first One sensor in The measured concentration value at time [time]. For the diffusion model in the leakage source parameter set The first calculated under the drive The theoretical concentration value at each sensor, To be compatible with sensors exist Weighting coefficients at time points, The length of the observation time window; S22: Set feasible region constraints for decision variables; set dynamic and static constraints in conjunction with plant production conditions: location of leakage source. Limited to potential leakage areas, eliminating invalid areas; leakage rate The lower limit is set as the critical leakage rate corresponding to the lowest detectable concentration of the sensor, and the upper limit is the maximum theoretical leakage rate under the rated pressure of the equipment; leakage initiation time No earlier than a preset time period preceding the first time the sensor detects an anomaly; type of leaked substance Limited to the types of gases actually present within the factory area; S23: Construct a source tracing and inversion optimization mathematical model; formalize the leakage source tracing into a constrained nonlinear optimization problem: ; In the formula, For the optimal set of leakage source parameters, and These are the lower and upper bounds for each decision variable, respectively.

[0010] As a preferred option, the hierarchical optimization solution strategy in step 3 is as follows: S31: Coarse Search Stage - Rapid Location of Leak Source Area; Particle Swarm Optimization (PSO) or Genetic Algorithm (GA) is used for global search; a candidate solution population is randomly generated within the feasible region, using the potential leak area of ​​the plant as a constraint; candidate parameters are... Input diffusion model to solve for concentration field, using fitness The evaluation process is performed; the population is iteratively updated until the convergence threshold is met or the maximum number of coarse search iterations is reached, and local candidate regions containing real leakage sources are locked; when the preset convergence condition is met, the process automatically switches to the fine search stage. S32: Fine Search Stage - Fine Inversion of Leakage Source Parameters; Using the optimal candidate solution from the coarse search as the initial value, local optimization is carried out using gradient descent, Levenberg-Marquardt algorithm, or simplex method; a regularization term is introduced into the objective function to constrain the variation of parameters in adjacent iterations and suppress solution oscillations caused by observation noise; an upper limit for fine search iterations and convergence criteria are set. S33: Uncertainty quantification; obtaining the optimal solution Then, using the objective function The system estimates the posterior confidence intervals of each parameter in the local curvature estimation of the neighborhood of the optimal solution; it outputs a confidence ellipse at the location of the leakage source to visually represent the location uncertainty; when the area of ​​the confidence ellipse exceeds a preset threshold, the system automatically pushes a prompt, suggesting the deployment of temporary measurement points or manual on-site verification; if the nonlinearity of the objective function is too high and the curvature method confidence estimation fails, the Monte Carlo sampling method is used to assist in assessing the uncertainty.

[0011] As a preferred option, in step 4, the process of outputting leakage source parameters and performing online verification and adaptive correction is as follows: S41: Output the optimal leakage source parameter set Simultaneously output the position confidence ellipse, confidence intervals for each parameter, and optimized convergence curve; S42: Online verification; as the leak continues, new sensor observation data is constantly generated; optimizing the leak source parameter set. Substitute the values ​​into the diffusion model to calculate the predicted concentration values ​​for future periods. , and subsequent real-time observations Compare the predicted and observed values; if the deviation between the predicted and observed values ​​continues to increase and exceeds the preset threshold, it indicates that there may be a deviation in the source tracing results, triggering adaptive correction: using the current optimal solution as the initial value, re-execute the inversion optimization of the fine search stage using the newly added observation data, and update the leakage source parameter estimates. S43: Multi-leakage source identification; set a residual judgment threshold. If the optimal objective function value of a single leakage source model is continuously higher than the threshold, the multi-leakage source identification mode is activated; group or time segment independent inversion of the observation data to distinguish the location and parameters of multiple leakage sources; consider the superposition interference of multi-source diffusion plumes in the inference process to reduce the probability of missed identification and repeated identification.

[0012] As a preferred option, in step 5, the emergency response linkage process driven by the source tracing results is as follows: S51: Mark the coordinates of the leak source location obtained from the source tracing on the plant's GIS map as the core reference point for emergency response; quickly generate a simplified hazard diffusion outline map centered on the leak source location and combined with real-time wind direction and speed. S52: Based on the location of the leak source, the leakage rate, and the type of substance, it automatically determines the emergency response level and plans personnel evacuation routes radiating outward from the area surrounding the leak source; the evacuation routes automatically avoid high-risk diffusion areas centered on the leak source and guide people to safe areas in the crosswind or upwind direction. S53: Based on the type of leaked substance and the type of equipment where the leak source is located, the optimal handling measures are matched from the contingency plan library to generate a handling plan that includes specific operations such as shutting off valves to cut off material supply, starting sprinklers, and setting up area warnings; the handling plan is bound to the location of the leak source to guide emergency personnel to quickly and accurately reach the target equipment for handling.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention adopts a technical approach based on physical inversion and optimization, formalizing the leakage source tracing problem as a nonlinear optimization problem with leakage source parameters as decision variables, minimizing the weighted deviation between sensor-observed concentrations and diffusion model-theoretical concentrations as objectives, and constrained by the physical feasible region. This problem is solved through a hierarchical optimization strategy. This approach does not rely on a large amount of historical leakage case data for model training; it only requires real-time sensor observation data and a diffusion physics model to complete source tracing and inversion. This overcomes the limitations of data-driven deep learning methods in scenarios with insufficient training data, such as newly commissioned plants or rare leakage conditions.

[0014] 2. This invention constructs an observation-model weighted bias function for source tracing, using the measured concentration of each sensor as the weight coefficient. This allows high-concentration sensors closer to the leak source to play a greater role in optimization, improving the sensitivity of source tracing to the location of the leak source. Simultaneously, by setting constraints on decision variables based on the physical feasible region—limiting the leak source location to potential leak areas such as equipment and pipelines, limiting the leak rate to the range allowed by the equipment's physical parameters, and constraining the leak initiation time to near the moment the sensor first detects the anomaly—prior physical knowledge of the plant area is explicitly embedded into the optimization problem, effectively reducing the search space and improving the efficiency and stability of the inversion solution.

[0015] 3. This invention employs a hierarchical optimization strategy of coarse search followed by fine search. In the coarse search stage, particle swarm optimization or genetic algorithms are used to quickly locate candidate leakage source regions within the global feasible region. In the fine search stage, gradient descent or the Levenberg-Marquardt algorithm is used to finely invert leakage source parameters within a local region. This hierarchical strategy balances the timeliness requirements of emergency response with the accuracy requirements of source tracing by switching between sparse and dense computational grids and between a small number and a large number of diffusing particles. The fine search stage introduces a regularization term to constrain the variation amplitude of parameters in adjacent iterations, effectively suppressing solution oscillations caused by observation noise.

[0016] 4. This invention introduces an uncertainty quantification mechanism, utilizing the local curvature information of the objective function in the neighborhood of the optimal solution to estimate the posterior confidence interval of each parameter, and outputting a confidence ellipse at the location of the leakage source to intuitively represent the location uncertainty. When the area of ​​the confidence ellipse exceeds a preset threshold, it automatically prompts the need to add temporary measurement points or conduct manual verification. This mechanism provides a quantifiable reliability evaluation for the source tracing results, overcoming the limitation of traditional optimization methods that only output point estimates and lack confidence references.

[0017] 5. This invention establishes an online verification and adaptive correction mechanism, continuously validating the source tracing results using newly added observation data. When the normalized root mean square deviation between the predicted and measured concentrations shows a monotonically increasing trend over multiple consecutive monitoring periods and exceeds a preset threshold, the inversion optimization update of the leakage source parameter estimation during the fine search phase is automatically triggered. This mechanism enables the source tracing results to be continuously optimized with data accumulation, while also supporting the identification of multiple leakage sources and handling the identification problem of multi-source diffusion plumes superimposed in complex leakage scenarios.

[0018] 6. This invention achieves adaptive selection of diffusion models through a wind field stability quantification index. This index is obtained by weighted fusion of wind direction stability components and wind speed stability components. When the index exceeds a preset threshold, the Gaussian diffusion model is automatically selected, and when it is below the threshold, the Lagrange particle diffusion model is automatically selected, so that the forward concentration calculation can automatically achieve a balance between calculation speed and simulation accuracy according to real-time meteorological conditions.

[0019] This method transforms the source tracing of metallurgical gas leaks from experience-based judgments relying on historical data into quantitative inversion driven by real-time observation and physical models, achieving precise source location and emergency decision-making loop under zero-sample conditions.

[0020] This invention also provides a metallurgical gas leak tracing system based on physical inversion and optimization, used to implement a metallurgical gas leak tracing method based on physical inversion and optimization, including: The plant area physical model construction module is used to acquire plant area geographic information, meteorological data and sensor layout information, complete data preprocessing and time series alignment, and construct a three-dimensional diffusion physical model of the plant area for leak tracing. The source tracing and inversion optimization problem modeling module is used to establish a constrained source tracing and inversion nonlinear optimization problem with leakage source parameters as decision variables and minimizing the deviation between the concentration observed by the sensor and the theoretical concentration of the diffusion model as the objective. The hierarchical optimization solution module is used to solve for the optimal leakage source parameter set by adopting a two-level progressive optimization strategy of first coarse search and then fine search, and to quantify the uncertainty of the source tracing results; The online verification and adaptive correction module is used to continuously verify the source tracing results using newly added observation data and trigger adaptive correction when the deviation exceeds the limit. It also supports the identification of multiple leakage sources. The source tracing-driven emergency response module is used to automatically generate graded emergency plans based on the leak source location and leak parameters obtained from source tracing, and supports closed-loop update of simulation results after handling.

[0021] As a preferred embodiment, the plant area physical model construction module includes a 3D geometric modeling unit, a sensor network deployment unit, a 3D wind field solving unit, and a diffusion model construction unit. The 3D geometric modeling unit acquires geographic information data such as plant topography, building distribution, and geometric parameters of structures through a GIS system, and quickly assembles and generates a 3D geometric model of the plant area by calling the system's built-in metallurgical standard component library, marking leak-free areas and potential leaking equipment pipe sections within the plant area. The sensor network deployment unit collects the 3D coordinates of all gas sensor installation locations, sensor models, measurement accuracy, and saturation threshold parameters, marking the sensor locations in the 3D geometric model to form sensor observation data. The network synchronously acquires real-time meteorological datasets and unifies the timestamps of all monitoring data to complete time-series registration. Simultaneously, it preprocesses sensor data, removing abnormal measurement point data due to signal drift, numerical saturation, or sudden failure. The three-dimensional wind field solution unit uses real-time meteorological data as boundary conditions and employs a fast wind field diagnostic model or a pre-trained surrogate wind field model to solve for the three-dimensional spatial wind field distribution within the factory area. It considers building obstruction and airflow around effects and sets a building density discrimination criterion to automatically select the wind field model. The diffusion model construction unit automatically selects a forward diffusion model based on the wind field stability quantification index, using the leakage source parameter set as input and outputting the theoretical concentration values ​​at each sensor location. The source tracing and inversion optimization problem modeling module includes an observation-model bias function construction unit and a decision variable feasible region constraint setting unit. The observation-model bias function construction unit establishes a weighted bias objective function with the sensor's measured concentration as the weight coefficient, implements weight suppression for sensors that reach the saturation threshold, and introduces a sensor's inherent measurement error normalization term to take into account the accuracy differences of different sensor models. The decision variable feasible region constraint setting unit limits the location of the leak source to potential leak areas such as equipment and pipelines, sets the lower limit of the leak rate to the critical leak rate and the upper limit to the maximum theoretical leak rate under the equipment's rated pressure, constrains the leak initiation time to near the moment when the sensor first detects the anomaly, limits the leaked substance type to the types of gases actually present in the plant area, and can dynamically update the feasible region range according to the equipment's start-up, shutdown, and maintenance status. This module ultimately formalizes the leak source tracing problem into a constrained nonlinear optimization mathematical model. The hierarchical optimization solution module includes a coarse search unit, a fine search unit, and an uncertainty quantification unit. The coarse search unit uses a particle swarm optimization algorithm or a genetic algorithm to search within the global feasible region. It uses the potential leakage area of ​​the plant as a constraint and employs a sparse computational grid and a small number of particles to quickly identify local candidate regions containing the actual leakage source. The fine search unit uses the optimal candidate solution from the coarse search as the initial value and employs gradient descent, Levenberg-Marquardt algorithm, or simplex method for local optimization. It uses a denser grid and a larger number of particles to improve the accuracy of concentration calculation. A regularization term is introduced into the objective function to constrain the variation of parameters in adjacent iterations to suppress solution oscillations caused by observation noise. The uncertainty quantification unit uses the local curvature of the objective function in the neighborhood of the optimal solution to estimate the posterior confidence interval of each parameter. It outputs a confidence ellipse at the leakage source location to visually represent the location uncertainty. When the area of ​​the confidence ellipse exceeds a preset threshold, it automatically pushes a prompt suggesting the deployment of temporary measurement points or manual on-site verification. When the nonlinearity of the objective function is too high, causing the curvature method confidence estimation to fail, Monte Carlo sampling is used to assist in evaluating the uncertainty.

[0022] As a preferred embodiment, the online verification and adaptive correction module includes an online verification unit, an adaptive correction unit, and a multi-leakage source identification unit. The online verification unit substitutes the optimal leakage source parameter set into the diffusion model to calculate the predicted concentration value for future periods, compares it with subsequent real-time observations, sets a continuous monitoring window length, and calculates the normalized root mean square deviation between the predicted and measured concentrations. When this deviation shows a monotonically increasing trend over multiple consecutive monitoring periods and exceeds a preset threshold, it is determined that the source tracing result has a significant deviation. When the adaptive correction unit determines that the source tracing result has a deviation, it uses the current optimal solution as the initial value and re-executes a fine search optimization using newly added observation data to update the leakage source parameter estimates. The multi-leakage source identification unit sets a residual judgment threshold. When the optimal objective function value of a single leakage source model is continuously higher than the threshold, it activates the multi-leakage source identification mode, groups or independently inverts the observation data in time segments, and distinguishes the locations and parameters of multiple leakage sources. As a preferred embodiment, the source tracing-driven emergency response module includes a hazard distribution generation unit, an evacuation route planning unit, and a disposal plan matching unit. The hazard distribution generation unit marks the coordinates of the leak source location on the plant's GIS map and quickly generates a rough distribution map of hazard diffusion centered on the leak source, combined with real-time wind direction and speed. The evacuation route planning unit automatically determines the emergency response level based on the leak source location, leakage rate, and substance type, and plans personnel evacuation routes radiating outward from the area surrounding the leak source. The evacuation routes automatically avoid high-risk diffusion areas centered on the leak source and incorporate constraints from the plant's fire lanes, work zones, and fixed post distribution, guiding personnel towards safe areas in the crosswind or upwind direction. The disposal plan matching unit matches the optimal disposal measures from the contingency plan library based on the leaked substance type and the type of equipment where the leak source is located, generating a disposal plan that includes specific operations such as valve shut-off, sprinkler activation, and area warning. The disposal plan is bound to the leak source location to guide emergency personnel to quickly and accurately reach the target equipment for disposal.

[0023] In this invention, the plant area physical model construction module transforms general geographic information data into a dedicated physical model for leak tracing, providing precise geometric boundaries and spatial constraints for subsequent diffusion calculations and source tracing inversion. The source tracing inversion optimization problem modeling module transforms leak tracing from traditional empirical judgment into a quantifiable mathematical optimization problem. The hierarchical optimization solution module achieves a dynamic balance between computational speed and positioning accuracy through a two-stage progressive strategy of coarse search followed by fine search. The online verification and adaptive correction module continuously verifies and corrects the tracing results by incorporating newly added sensor observation data, ensuring that tracing accuracy continuously improves with data accumulation. Finally, the source tracing-driven emergency response module directly converts the leak source location and release parameters obtained from the tracing into executable emergency response decisions.

[0024] This system uses multi-field coupled physical modeling to recreate the actual diffusion process in the plant area, and achieves stable and reliable leakage source location through hierarchical optimization and uncertainty quantification. The source tracing results are then directly mapped to hierarchical emergency operation plans. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0026] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0027] like Figure 1As shown, this invention provides a method for tracing the source of metallurgical gas leaks based on physical inversion and optimization solutions, including the following steps: Step 1: Construct a three-dimensional diffusion physical model of the plant area; acquire geographical information, meteorological data and sensor layout information of the plant area, complete data preprocessing and time series alignment, and construct a three-dimensional diffusion physical model of the plant area for leak tracing. As a preferred option, the process of constructing a three-dimensional diffusion physical model of the plant area is as follows: S11: Obtain geographic information data such as plant topography, building distribution, and geometric parameters of structures through the GIS system; call the system's built-in metallurgical standard component library (including parametric models of blast furnaces, converters, gas holders, pipe corridors, and coke ovens) to quickly assemble and generate a three-dimensional geometric model of the plant area, and mark the plant area's leak-prohibited areas and potential leak equipment pipe sections. S12: Collect the three-dimensional coordinates of all gas sensor installation locations, sensor models, measurement accuracy, and saturation threshold parameters; mark the sensor locations in the three-dimensional geometric model to form a sensor observation network; synchronously acquire real-time meteorological datasets, including wind speed, wind direction, temperature, humidity, and atmospheric stability level, and unify the timestamps of all monitoring data to complete time-series registration; As a preferred method, the sensor pre-data undergoes the following operations sequentially: abnormal measurement point data due to signal drift, numerical saturation, and abrupt failure are removed. Specifically, for signal drift, if the average concentration of a sensor within a preset time window deviates from the average concentration of the entire sensor network during the same period by more than a preset multiple (e.g., 3 standard deviations), it is considered signal drift. For numerical saturation, if the sensor output value continuously exceeds a preset proportion (e.g., 95%) of the rated range, it is considered saturation. For abrupt failure, if the concentration change rate of adjacent sampling points exceeds the physically possible maximum diffusion rate, it is considered abrupt failure. Sensor data deemed abnormal within this time window does not participate in the subsequent deviation function calculation, and the corresponding weighting coefficient... Set to zero.

[0028] S13: Construct a three-dimensional wind field in the plant area; using real-time meteorological data as boundary conditions, solve for the three-dimensional spatial wind field distribution in the plant area using a fast wind field diagnostic model or a pre-trained surrogate wind field model. The effects of building obstruction and airflow around the building are taken into account; a building density discrimination criterion is set to realize the automatic selection of wind field model: logarithmic wind profile is used for open areas, and a fast wind field parameterization scheme based on the simplified solution of RANS equation is used for densely built areas. S14: Construct a diffusion physical model; automatically select a forward diffusion model based on wind field stability quantification indicators; the wind field stability quantification indicators Wind direction stability component and wind speed stability component The weighted fusion yields: ,in , , and These represent the standard deviations of wind direction and wind speed within the preset time window. and For the preset threshold, For weighting coefficients; when When the wind field is considered stable, the Gaussian diffusion model is selected; when... When the wind field is determined to be non-stationary, the Lagrange particle diffusion model is selected; the diffusion model uses the leakage source parameter set. As input, output the theoretical concentration value at each sensor location. ,in The three-dimensional coordinates of the leakage source For leakage rate, The leak start time, Type of leaked substance; In this technical solution, a three-dimensional geometric model of the plant area is quickly assembled by calling a metallurgical standard component library and potential leakage areas are marked, providing precise spatial constraint boundaries for source tracing and inversion. By constructing a sensor observation network covering the entire plant area, meteorological data is synchronized to complete time-series registration, and quantitative elimination criteria are set for abnormal measurement point data according to three categories: signal drift, numerical saturation, and abrupt failure, ensuring the reliability of subsequent deviation function calculations from the data source. The wind field model is automatically selected through the building density discrimination criterion, achieving an adaptive balance between computational efficiency and accuracy. The wind field stability quantification index obtained by weighted fusion of wind direction stability and wind speed stability enables the adaptive selection of Gaussian model and Lagrange particle model, allowing the forward concentration calculation to automatically switch between computational speed and simulation accuracy according to real-time meteorological conditions, providing a physical model foundation that balances timeliness and accuracy for hierarchical optimization solutions.

[0029] Step 2: Construct a source tracing and inversion optimization problem; using the leakage source parameters as decision variables and minimizing the deviation between the concentration observed by the sensor and the theoretical concentration of the diffusion model as the objective, establish a constrained source tracing and inversion optimization problem; As a preferred approach, the process of constructing the source tracing and inversion optimization problem is as follows: S21: Establish the observation-model bias function; for those with A network of gas sensors is used to define a weighted deviation function between the observed concentration and the theoretical concentration. : ; In the formula, For the first One sensor in The measured concentration value at time [time]. For the diffusion model in the leakage source parameter set The first calculated under the drive The theoretical concentration value at each sensor, To be compatible with sensors exist Weighting coefficients at time points, The length of the observation time window is defined. The weights are positively correlated with the measured concentrations, and upper and lower limits are added. Weight suppression is implemented for sensors that have reached the saturation threshold. The weights incorporate a normalization term for the inherent measurement error of the sensor, taking into account the accuracy differences between different sensor models. S22: Set feasible region constraints for decision variables; set dynamic and static constraints in conjunction with plant production conditions: location of leakage source. Limited to potential leakage areas such as equipment and pipelines, excluding ineffective areas such as open spaces and building rooftops; leakage rate The lower limit is set as the critical leakage rate corresponding to the lowest detectable concentration of the sensor, and the upper limit is the maximum theoretical leakage rate under the rated pressure of the equipment; leakage initiation time No earlier than a preset time period preceding the first time the sensor detects an anomaly; type of leaked substance Limited to the types of gases actually present in the plant area; as an optimal option, the feasible range can be dynamically updated based on the equipment start-up, shutdown, and maintenance status; Critical leakage rate The minimum detectable concentration of the sensor is determined as follows: Substituting into the forward diffusion model, assuming the leak source is located in the nearest potential leak area of ​​the sensor, and using the average wind speed and neutral atmospheric stability level under current meteorological conditions, the inverse solution satisfies... Minimum required leakage rate The critical leakage rate is used as the reference value. To reduce the amount of online calculation, the critical leakage rate of each sensor can be pre-calculated and stored as a lookup table based on the sensor layout in the plant area, typical meteorological conditions and common leakage media. The sensor can then be directly retrieved from the table during online operation.

[0030] S23: Construct a source tracing and inversion optimization mathematical model; formalize the leakage source tracing into a constrained nonlinear optimization problem: ; In the formula, For the optimal set of leakage source parameters, and These are the lower and upper bounds for each decision variable, respectively. In this technical solution, a weighted deviation objective function with sensor-measured concentrations as weights is established to quantify the deviation between multi-sensor time-series observation data and the theoretical values ​​of the diffusion model into a unified optimization objective. By implementing weight suppression for saturated sensors and introducing a normalization term for inherent sensor measurement errors, the accuracy differences of different sensor models are taken into account, improving the robustness of the source tracing solution. By setting decision variable constraints based on the physical feasible region—limiting the leak source location to the potential leak area, limiting the leak rate to the range of equipment physical parameters, and constraining the leak initiation time to near the moment when the sensor first detects the anomaly—prior physical knowledge of the plant area is explicitly embedded into the optimization problem, effectively reducing the search space and improving the efficiency and stability of the inversion solution. Simultaneously, the critical leak rate is obtained by substituting the lowest detected concentration of the sensor into the forward diffusion model for inverse solution or by using a pre-calculated lookup table, balancing computational accuracy and online operating efficiency. Finally, the leak source tracing is formalized into a constrained nonlinear optimization mathematical model, transforming the source tracing process from an experience-based qualitative judgment into a quantifiable mathematical problem.

[0031] Step 3: Hierarchical optimization solution strategy; adopt a two-level progressive optimization strategy of first coarse search and then fine search, and add a global maximum computation time constraint to balance the timeliness of emergency response and the accuracy of source tracing; As a preferred approach, the hierarchical optimization solution strategy proceeds as follows: S31: Coarse Search Stage - Rapid Location of Leak Source Area; Particle Swarm Optimization (PSO) or Genetic Algorithm (GA) is used for global search; a candidate solution population is randomly generated within the feasible region, using the potential leak area of ​​the plant as a constraint; candidate parameters are... Input diffusion model to solve for concentration field, using fitness The evaluation process is performed; the population is iteratively updated until the convergence threshold is met or the maximum number of coarse search iterations is reached, and local candidate regions containing real leakage sources are locked; the coarse search uses a sparse computing grid and a smaller number of particles to shorten the computation time; the process automatically switches to the fine search stage when the preset convergence condition is met. S32: Fine Search Stage - Refined Inversion of Leakage Source Parameters; Using the optimal candidate solution from the coarse search as the initial value, local optimization is carried out using gradient descent, Levenberg-Marquardt algorithm, or simplex method; the fine search uses a denser grid and a larger number of particles to improve the accuracy of concentration calculation; a regularization term is introduced into the objective function to constrain the variation of parameters in adjacent iterations and suppress solution oscillations caused by observation noise; an upper limit for the fine search iteration and convergence criteria are set; As a preferred option, the regularization term coefficient The value range is related to the observed noise level and the concentration calculation error of the diffusion model. Under typical signal-to-noise ratio conditions (sensor measurement error not exceeding 5% of full scale), Desirable Up to 10 The values ​​between; in scenarios with high sensor noise or high uncertainty in the diffusion model. It can be appropriately increased to 10 Up to 10 The specific value can be optimized and adjusted based on the cross-validation results of historical source tracing cases.

[0032] S33: Uncertainty quantification; obtaining the optimal solution Then, using the objective function The system estimates the posterior confidence intervals of each parameter in the local curvature estimation of the neighborhood of the optimal solution; it outputs the confidence ellipse of the leakage source location to intuitively represent the location uncertainty; when the area of ​​the confidence ellipse exceeds the preset threshold, the system automatically pushes a prompt and suggests setting up temporary measurement points or conducting manual on-site verification; if the nonlinearity of the objective function is too high and the curvature method confidence estimation fails, the Monte Carlo sampling method is used to assist in evaluating the uncertainty. This technical solution achieves a dynamic balance between computational speed and positioning accuracy through a two-stage progressive optimization strategy of coarse search followed by fine search. In the coarse search stage, a particle swarm optimization algorithm or genetic algorithm is used for rapid searching within the global feasible region. Combined with a sparse computational grid and a smaller number of diffusing particles, this enables the rapid identification of candidate leak source regions after a leak occurs, meeting the timeliness requirements of emergency response. In the fine search stage, the optimal solution from the coarse search is used as the initial value, and a gradient descent method or Levenberg-Marquardt algorithm is employed for refined inversion within a local region. A denser grid and a larger number of diffusing particles further improve the accuracy of concentration calculation. By introducing a regularization term into the objective function to constrain the variation amplitude of parameters in adjacent iterations, solution oscillations caused by observation noise are effectively suppressed, improving the stability and accuracy of source tracing parameter estimation. Simultaneously, the posterior confidence intervals of each parameter are estimated using the local curvature information of the objective function in the neighborhood of the optimal solution. The output leak source location confidence ellipse visually represents the positioning uncertainty. When the curvature method fails, Monte Carlo sampling is used for auxiliary evaluation, providing a quantifiable reliability assessment of the source tracing results.

[0033] Step 4: Output and online verification of leakage source parameters and adaptive correction; output the optimal leakage source parameters, continuously verify the source tracing results using newly added observation data, and support the identification of multiple leakage sources; As a preferred option, the process of outputting leakage source parameters and performing online verification and adaptive correction is as follows: S41: Output the optimal leakage source parameter set It synchronously outputs the location confidence ellipse, the confidence interval of each parameter, and the optimized convergence curve; it performs the simulation by default with a constant leakage rate; for time-varying leakage scenarios, it can be extended to a time-varying leakage rate sequence optimization solution mode. S42: Online verification; as the leak continues, new sensor observation data is constantly generated; optimizing the leak source parameter set. Substitute the values ​​into the diffusion model to calculate the predicted concentration values ​​for future periods. , and subsequent real-time observations Compare the predicted and observed values; if the deviation between the predicted and observed values ​​continues to increase and exceeds the preset threshold, it indicates that there may be a deviation in the source tracing results, triggering adaptive correction: using the current optimal solution as the initial value, re-execute the inversion optimization of the fine search stage using the newly added observation data, and update the leakage source parameter estimates. Set the continuous monitoring window length to (If 3 to 5 sampling periods are used), calculate the normalized root mean square deviation (NRMSD) between the predicted concentration and the measured concentration in each monitoring period. in This represents the average measured concentration from all valid sensors during the same period. When NRMSD is in continuous... If the NRMSD of the predicted value and the observed value show a monotonically increasing trend within a monitoring period, and the NRMSD of the most recent period exceeds the preset threshold (e.g., 0.3 to 0.5), then it is determined that the deviation between the predicted value and the observed value continues to increase and exceeds the preset threshold, indicating a significant deviation in the source tracing results, thus triggering the adaptive correction process.

[0034] S43: Multi-leakage source identification; Set a residual judgment threshold. If the optimal objective function value of a single leakage source model is consistently higher than the threshold, the multi-leakage source identification mode is activated; Group or time-segmented independent inversion of the observation data to distinguish the locations and parameters of multiple leakage sources; The inference process considers the superposition interference of multi-source diffusion plumes to reduce the probability of missed identification and duplicate identification. This technical solution constructs a closed-loop adaptive source tracing mechanism encompassing output, verification, and correction. It not only outputs the optimal leakage source parameters and their uncertainty quantification results, such as confidence intervals and confidence ellipses, but also performs online verification by continuously monitoring the normalized root mean square deviation (NRMSD) between predicted and measured concentrations within a continuous monitoring window. When the NRMSD exhibits a monotonically increasing trend over multiple consecutive monitoring periods and exceeds a preset threshold, it automatically triggers a fine-search phase inversion optimization with the current solution as the initial value, achieving adaptive updates to the leakage source parameters. Simultaneously, a residual judgment threshold is set. When the objective function value of the single-source model remains consistently high, a multi-leakage source identification mode is automatically activated. Through grouping or time-segmented independent inversion of observed data, multiple leakage sources are distinguished under the condition of considering the superposition interference of multi-source diffusion plumes. This effectively solves the problems of static solidification of source tracing results, inability to dynamically correct with data accumulation, and leakage and duplicate identification in complex multi-leakage scenarios in traditional methods.

[0035] Step 5: Emergency response linkage driven by source tracing results; Based on the location and parameters of the leak source obtained from the source tracing, an emergency response decision centered on the location of the leak source is automatically triggered; As a preferred approach, the emergency response coordination process driven by the source tracing results is as follows: S51: Mark the coordinates of the leak source location obtained from the source tracing on the plant's GIS map as the core reference point for emergency response; quickly generate a simplified hazard diffusion outline map centered on the leak source location and combined with real-time wind direction and speed. S52: Based on the location of the leak source, the leakage rate, and the type of substance, it automatically determines the emergency response level and plans personnel evacuation routes radiating outward from the area surrounding the leak source; the evacuation routes automatically avoid high-risk diffusion areas centered on the leak source and guide people to safe areas in the crosswind or upwind direction. S53: Based on the type of leaked substance and the type of equipment where the leak source is located, the optimal handling measures are matched from the contingency plan library to generate a handling plan that includes specific operations such as shutting off valves to cut off material supply, starting sprinklers, and setting up area warnings; the handling plan is bound to the location of the leak source to guide emergency personnel to quickly and accurately reach the target equipment for handling.

[0036] This technical solution constructs a direct, closed-loop linkage from source tracing results to emergency decision-making, transforming the leak source parameters located through inversion into actionable emergency response actions in real time. By precisely marking the leak source location coordinates on the plant's GIS map and coupling them with real-time wind fields to generate a hazard diffusion overview map, it solves the problems of unclear leak source location and diffusion situation in traditional emergency responses. Based on the leak source location, rate, and substance type, it automatically determines the response level and plans personnel evacuation routes centered on the leak source, automatically avoiding high-risk diffusion areas, and guided towards the crosswind or upwind direction, thus upgrading evacuation decision-making from experience-based judgment to quantitative guidance. Simultaneously, based on the leaked substance type and equipment type, it matches specific handling plans from the contingency plan database, including valve shut-off, sprinkler activation, and area warning, and the handling plans are bound to the leak source location, enabling emergency personnel to quickly and accurately reach the target equipment to carry out the handling, effectively shortening the response chain from monitoring and alarm to precise handling.

[0037] This invention also provides a metallurgical gas leak tracing system based on physical inversion and optimization, used to implement a metallurgical gas leak tracing method based on physical inversion and optimization, including: The plant area physical model construction module is used to acquire geographical information, meteorological data, and sensor layout information of the plant area, complete data preprocessing and time-series alignment, and construct a three-dimensional diffusion physical model of the plant area for leak tracing. The module includes a three-dimensional geometric modeling unit, a sensor network deployment unit, a three-dimensional wind field solving unit, and a diffusion model construction unit. The three-dimensional geometric modeling unit acquires geographical information data such as plant topography, building distribution, and geometric parameters of structures through a GIS system, and quickly assembles and generates a three-dimensional geometric model of the plant area by calling the system's built-in metallurgical standard component library, marking leak-prohibited areas and potential leaking equipment pipe sections. The sensor network deployment unit collects the three-dimensional coordinates of all gas sensor installation locations, sensor models, measurement accuracy, and saturation values. Threshold parameters are used to mark sensor locations in a 3D geometric model, forming a sensor observation network. Real-time meteorological datasets are acquired synchronously, and timestamps of all monitoring data are unified for time-series registration. Simultaneously, sensor data is preprocessed to remove abnormal measurement point data due to signal drift, numerical saturation, or sudden failure. The 3D wind field solution unit uses real-time meteorological data as boundary conditions and employs a fast wind field diagnostic model or a pre-trained surrogate wind field model to solve for the 3D spatial wind field distribution within the plant area. Building obstruction and airflow around the source are taken into account, and a building density discrimination criterion is set to automatically select the wind field model. The diffusion model construction unit automatically selects a forward diffusion model based on wind field stability quantification indicators, using the leakage source parameter set as input and outputting the theoretical concentration values ​​at each sensor location. The source tracing and inversion optimization problem modeling module is used to establish a constrained source tracing and inversion nonlinear optimization problem with leakage source parameters as decision variables and the objective of minimizing the deviation between the sensor-observed concentration and the theoretical concentration of the diffusion model. This module includes an observation-model deviation function construction unit and a decision variable feasible region constraint setting unit. The observation-model deviation function construction unit establishes a weighted deviation objective function with the sensor's measured concentration as the weight coefficient, implements weight suppression for sensors that reach the saturation threshold, and introduces a normalization term for the inherent measurement error of the sensor to account for the accuracy differences between different sensor models. The decision variable feasible region constraint setting unit limits the leakage source location to potential leakage areas such as equipment and pipelines, sets the lower limit of the leakage rate to the critical leakage rate and the upper limit to the maximum theoretical leakage rate under the rated pressure of the equipment, constrains the leakage initiation time to near the moment when the sensor first detects an anomaly, and limits the leakage substance type to the types of gases actually present in the plant area. It can also dynamically update the feasible region range based on the equipment start-up, shutdown, and maintenance status. This module ultimately formalizes the leakage source tracing problem into a constrained nonlinear optimization mathematical model. The hierarchical optimization solution module is used to solve for the optimal leakage source parameter set using a two-level progressive optimization strategy of coarse search followed by fine search, and to quantify the uncertainty of the source tracing results. The hierarchical optimization solution module includes a coarse search unit, a fine search unit, and an uncertainty quantification unit. The coarse search unit uses a particle swarm optimization algorithm or a genetic algorithm to search within the global feasible region, using the potential leakage area of ​​the plant as a constraint, and employs a sparse computational grid and a small number of particles to quickly identify local candidate regions containing the actual leakage source. The fine search unit uses the optimal candidate solution from the coarse search as the initial value and employs gradient descent, the Levenberg-Marquardt algorithm, or... The simplex method is used for local optimization, and a denser grid and a larger number of particles are used to improve the accuracy of concentration calculation. A regularization term is introduced into the objective function to constrain the variation of parameters in adjacent iterations to suppress solution oscillations caused by observation noise. The uncertainty quantification unit uses the local curvature of the objective function in the neighborhood of the optimal solution to estimate the posterior confidence interval of each parameter, and outputs a confidence ellipse of the leakage source location to intuitively represent the location uncertainty. When the area of ​​the confidence ellipse exceeds a preset threshold, a prompt is automatically pushed suggesting the deployment of temporary measurement points or manual on-site verification. When the nonlinearity of the objective function is too high, causing the curvature method confidence estimation to fail, a Monte Carlo sampling method is used to assist in the evaluation of uncertainty. The online verification and adaptive correction module is used to continuously verify the source tracing results using newly added observation data and trigger adaptive correction when the deviation exceeds the limit. It also supports the identification of multiple leakage sources. The online verification and adaptive correction module includes an online verification unit, an adaptive correction unit, and a multiple leakage source identification unit. The online verification unit substitutes the optimal leakage source parameter set into the diffusion model to calculate the concentration prediction value for future periods, compares it with the subsequent real-time observation values, sets the continuous monitoring window length, and calculates the normalized root mean square deviation between the predicted concentration and the measured concentration. When the deviation shows a monotonically increasing trend over multiple consecutive monitoring periods and exceeds a preset threshold, it is determined that there is a significant deviation in the source tracing results. When the adaptive correction unit determines that there is a deviation in the source tracing results, it uses the current optimal solution as the initial value and re-executes fine search optimization using newly added observation data to update the leakage source parameter estimates. The multiple leakage source identification unit sets a residual judgment threshold. When the optimal objective function value of a single leakage source model is continuously higher than the threshold, the multiple leakage source identification mode is activated. The observation data is grouped or independently inverted in time segments to distinguish the locations and parameters of multiple leakage sources. The source-tracing-driven emergency response module automatically generates tiered emergency plans based on the leak source location and leakage parameters obtained from source tracing, and supports closed-loop updates of simulation results after handling. The module includes a hazard distribution generation unit, an evacuation route planning unit, and a response plan matching unit. The hazard distribution generation unit marks the leak source location coordinates on the plant's GIS map and quickly generates a rough hazard diffusion distribution map centered on the leak source, combined with real-time wind direction and speed. The evacuation route planning unit automatically determines the emergency response level based on the leak source location, leakage rate, and substance type, and plans personnel evacuation routes radiating outward from the area surrounding the leak source. These routes automatically avoid high-risk diffusion areas centered on the leak source and incorporate constraints from plant fire lanes, work zones, and fixed post distribution, guiding personnel towards safe areas in the crosswind or upwind direction. The response plan matching unit matches the optimal response measures from the contingency plan library based on the leaked substance type and the type of equipment where the leak source is located, generating a response plan that includes specific operations such as valve shut-off, sprinkler activation, and area warning. The response plan is bound to the leak source location to guide emergency personnel to quickly and accurately reach the target equipment for handling.

[0038] This invention uses multi-field coupled physical modeling to recreate the actual diffusion process in the plant area, and achieves stable and reliable leakage source location through hierarchical optimization and uncertainty quantification. The source tracing results are then directly mapped to hierarchical emergency operation plans.

[0039] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for tracing the source of metallurgical gas leaks based on physical inversion and optimization solutions, characterized in that, Includes the following steps: Step 1: Construct a three-dimensional diffusion physical model of the plant area; acquire geographical information, meteorological data and sensor layout information of the plant area, complete data preprocessing and time series alignment, and construct a three-dimensional diffusion physical model of the plant area for leak tracing. Step 2: Construct a source tracing and inversion optimization problem; using the leakage source parameters as decision variables and minimizing the deviation between the concentration observed by the sensor and the theoretical concentration of the diffusion model as the objective, establish a constrained source tracing and inversion optimization problem; Step 3: Hierarchical optimization solution strategy; adopt a two-level progressive optimization strategy of coarse search followed by fine search, and add a global maximum computation time constraint to balance the timeliness of emergency response and the accuracy of source tracing; Step 4: Output and online verification of leakage source parameters and adaptive correction; output the optimal leakage source parameters and continuously verify the source tracing results using subsequent newly added observation data; Step 5: Emergency response linkage driven by source tracing results; Based on the location and parameters of the leak source obtained from the source tracing, an emergency response decision centered on the location of the leak source is automatically triggered.

2. The method for tracing the source of metallurgical gas leakage based on physical inversion and optimization solution as described in claim 1, characterized in that, In step 1, the process of constructing the three-dimensional diffusion physical model of the plant area is as follows: S11: Obtain geographic information data of the plant area through the GIS system; call the system's built-in metallurgical standard component library to generate a three-dimensional geometric model of the plant area, and mark the leak-proof areas and potential leak equipment pipe sections in the plant area; S12: Collect the three-dimensional coordinates of all gas sensor installation locations, sensor models, measurement accuracy, and saturation threshold parameters; mark the sensor locations in the three-dimensional geometric model to form a sensor observation network; synchronously acquire real-time meteorological datasets, including wind speed, wind direction, temperature, humidity, and atmospheric stability level, and unify the timestamps of all monitoring data to complete time-series registration; S13: Construct a 3D wind field for the plant area; using real-time meteorological data as boundary conditions, solve for the 3D spatial wind field distribution of the plant area using a rapid wind field diagnostic model or a pre-trained surrogate wind field model. The effects of building obstruction and airflow around the building should be taken into account. The building density discrimination criterion is set to enable automatic selection of wind field model: logarithmic wind profile is used for open areas, and a fast wind field parameterization scheme based on simplified solution of RANS equation is used for densely built areas. S14: Construct a diffusion physical model; automatically select a forward diffusion model based on the wind field stability quantification index; the wind field stability quantification index Wind direction stability component and wind speed stability component The weighted fusion yields: ,in , , and These are the standard deviations of wind direction and wind speed within the preset time window, respectively. and For the preset threshold, For weighting coefficients; when When the wind field is considered stable, the Gaussian diffusion model is selected; when... When the wind field is determined to be non-stationary, the Lagrange particle diffusion model is selected; the diffusion model uses the leakage source parameter set. As input, output the theoretical concentration value at each sensor location. ,in The three-dimensional coordinates of the leakage source For leakage rate, The leak start time, The type of substance leaked.

3. The method for tracing the source of metallurgical gas leakage based on physical inversion and optimization solution as described in claim 1, characterized in that, In step 2, the process of constructing the source tracing and inversion optimization problem is as follows: S21: Establish the observation-model bias function; for those with A network of gas sensors is used to measure the concentration, and a weighted deviation function between the observed concentration and the theoretical concentration is defined. : ； In the formula, For the first One sensor in The measured concentration value at time [time]. For the diffusion model in the leakage source parameter set The first calculated under the drive The theoretical concentration value at each sensor, To be compatible with sensors exist Weighting coefficients at time points, The length of the observation time window; S22: Set feasible region constraints for decision variables; set dynamic and static constraints in conjunction with plant production conditions: location of leakage source. Limited to potential leakage areas, eliminating invalid areas; leakage rate The lower limit is set as the critical leakage rate corresponding to the lowest detectable concentration of the sensor, and the upper limit is the maximum theoretical leakage rate under the rated pressure of the equipment; leakage initiation time No earlier than a preset time period preceding the first time the sensor detects an anomaly; type of leaked substance Limited to the types of gases actually present within the factory area; S23: Construct a source tracing and inversion optimization mathematical model; formalize the leakage source tracing into a constrained nonlinear optimization problem: ； In the formula, For the optimal set of leakage source parameters, and These are the lower and upper bounds for each decision variable, respectively.

4. The method for tracing the source of metallurgical gas leakage based on physical inversion and optimization solution as described in claim 1, characterized in that, In step 3, the hierarchical optimization solution strategy proceeds as follows: S31: Coarse Search Stage - Rapid Location of Leak Source Area; Particle Swarm Optimization (PSO) or Genetic Algorithm (GA) is used for global search; a candidate solution population is randomly generated within the feasible region, using the potential leak area of ​​the plant as a constraint; candidate parameters are... Input diffusion model to solve for concentration field, using fitness The evaluation process is performed; the population is iteratively updated until the convergence threshold is met or the maximum number of coarse search iterations is reached, and local candidate regions containing real leakage sources are locked; when the preset convergence condition is met, the process automatically switches to the fine search stage. S32: Fine Search Stage - Fine Inversion of Leakage Source Parameters; Using the optimal candidate solution from the coarse search as the initial value, local optimization is carried out using gradient descent, Levenberg-Marquardt algorithm, or simplex method; a regularization term is introduced into the objective function to constrain the variation of parameters in adjacent iterations and suppress solution oscillations caused by observation noise; an upper limit for fine search iterations and convergence criteria are set. S33: Uncertainty quantification; obtaining the optimal solution Then, using the objective function The system estimates the posterior confidence intervals of each parameter in the local curvature estimation of the neighborhood of the optimal solution; it outputs a confidence ellipse at the location of the leakage source to visually represent the location uncertainty; when the area of ​​the confidence ellipse exceeds a preset threshold, the system automatically pushes a prompt, suggesting the deployment of temporary measurement points or manual on-site verification; if the nonlinearity of the objective function is too high and the curvature method confidence estimation fails, the Monte Carlo sampling method is used to assist in assessing the uncertainty.

5. The method for tracing the source of metallurgical gas leakage based on physical inversion and optimization solution as described in claim 1, characterized in that, In step 4, the process of outputting leakage source parameters and performing online verification and adaptive correction is as follows: S41: Output the optimal leakage source parameter set Simultaneously output the position confidence ellipse, confidence intervals for each parameter, and optimized convergence curve; S42: Online verification; as the leak continues, new sensor observation data is constantly generated; optimizing the leak source parameter set. Substitute the values ​​into the diffusion model to calculate the predicted concentration values ​​for future periods. , and subsequent real-time observations Perform a comparison; If the deviation between the predicted value and the observed value continues to increase and exceeds the preset threshold, it indicates that there may be a deviation in the source tracing results, triggering adaptive correction: using the current optimal solution as the initial value, the inversion optimization of the fine search stage is re-executed using the newly added observation data to update the leakage source parameter estimation; S43: Multi-leakage source identification; set a residual judgment threshold. If the optimal objective function value of a single leakage source model is continuously higher than the threshold, the multi-leakage source identification mode is activated; group or time segment independent inversion of the observation data to distinguish the location and parameters of multiple leakage sources; consider the superposition interference of multi-source diffusion plumes in the inference process to reduce the probability of missed identification and repeated identification.

6. The method for tracing the source of metallurgical gas leakage based on physical inversion and optimization solution as described in claim 1, characterized in that, In step 5, the emergency response linkage process driven by the source tracing results is as follows: S51: Mark the coordinates of the leak source location obtained from the source tracing on the plant's GIS map as the core reference point for emergency response; quickly generate a simplified hazard diffusion outline map centered on the leak source location and combined with real-time wind direction and speed. S52: Based on the location of the leak source, the leakage rate, and the type of substance, it automatically determines the emergency response level and plans personnel evacuation routes radiating outward from the area surrounding the leak source; the evacuation routes automatically avoid high-risk diffusion areas centered on the leak source and guide people to safe areas in the crosswind or upwind direction. S53: Based on the type of leaked substance and the type of equipment where the leak source is located, the optimal handling measures are matched from the contingency plan library to generate a handling plan that includes specific operations such as shutting off valves to cut off material supply, starting sprinklers, and setting up area warnings; the handling plan is bound to the location of the leak source to guide emergency personnel to quickly and accurately reach the target equipment for handling.

7. A metallurgical gas leak tracing system based on physical inversion and optimization, used to implement the metallurgical gas leak tracing method based on physical inversion and optimization as described in any one of claims 1 to 6, characterized in that, include: The plant area physical model construction module is used to acquire plant area geographic information, meteorological data and sensor layout information, complete data preprocessing and time series alignment, and construct a three-dimensional diffusion physical model of the plant area for leak tracing. The source tracing and inversion optimization problem modeling module is used to establish a constrained source tracing and inversion nonlinear optimization problem with leakage source parameters as decision variables and minimizing the deviation between the concentration observed by the sensor and the theoretical concentration of the diffusion model as the objective. The hierarchical optimization solution module is used to solve for the optimal leakage source parameter set by adopting a two-level progressive optimization strategy of first coarse search and then fine search, and to quantify the uncertainty of the source tracing results; The online verification and adaptive correction module is used to continuously verify the source tracing results using newly added observation data and trigger adaptive correction when the deviation exceeds the limit. It also supports the identification of multiple leakage sources. The source tracing-driven emergency response module is used to automatically generate graded emergency plans based on the leak source location and leak parameters obtained from source tracing, and supports closed-loop update of simulation results after handling.

8. The method for tracing the source of metallurgical gas leakage based on physical inversion and optimization solution according to claim 7, characterized in that, The plant area physical model construction module includes a 3D geometric modeling unit, a sensor network deployment unit, a 3D wind field solving unit, and a diffusion model construction unit. The 3D geometric modeling unit acquires geographic information data such as plant topography, building distribution, and geometric parameters of structures through a GIS system, and quickly assembles and generates a 3D geometric model of the plant area by calling the system's built-in metallurgical standard component library, marking leak-free areas and potential leak equipment pipe sections. The sensor network deployment unit collects the 3D coordinates of all gas sensor installation locations, sensor models, measurement accuracy, and saturation threshold parameters, marking the sensor locations in the 3D geometric model to form a sensor observation network. The system acquires real-time meteorological datasets and unifies the timestamps of all monitoring data to complete time-series registration. Simultaneously, it preprocesses sensor data to remove abnormal measurement point data due to signal drift, numerical saturation, or sudden failures. The three-dimensional wind field solution unit uses real-time meteorological data as boundary conditions and employs a fast wind field diagnostic model or a pre-trained surrogate wind field model to solve for the three-dimensional spatial wind field distribution within the factory area. It considers building obstruction and airflow around effects and sets a building density discrimination criterion to automatically select the wind field model. The diffusion model construction unit automatically selects a forward diffusion model based on the wind field stability quantification index, using the leakage source parameter set as input and outputting the theoretical concentration values ​​at each sensor location. The source tracing and inversion optimization problem modeling module includes an observation-model bias function construction unit and a decision variable feasible region constraint setting unit; The observation-model bias function construction unit establishes a weighted bias objective function with the sensor's measured concentration as the weight coefficient. It implements weighted suppression for sensors that reach the saturation threshold and introduces a normalization term for the inherent measurement error of the sensor to account for the accuracy differences between different sensor models. The decision variable feasible region constraint setting unit limits the location of the leak source to potential leak areas such as equipment and pipelines, sets the lower limit of the leak rate to the critical leak rate and the upper limit to the maximum theoretical leak rate under the rated pressure of the equipment, constrains the leak initiation time to near the moment the sensor first detects the anomaly, and limits the leaked substance type to the types of gases actually present in the plant area. It can also dynamically update the feasible region range based on the equipment's start-up, shutdown, and maintenance status. This module ultimately formalizes the leak tracing problem into a constrained nonlinear optimization mathematical model. The hierarchical optimization solution module includes a coarse search unit, a fine search unit, and an uncertainty quantification unit. The coarse search unit uses a particle swarm optimization algorithm or a genetic algorithm to search within the global feasible region. It uses the potential leakage area of ​​the plant as a constraint and employs a sparse computational grid and a small number of particles to quickly identify local candidate regions containing the actual leakage source. The fine search unit uses the optimal candidate solution from the coarse search as the initial value and employs gradient descent, Levenberg-Marquardt algorithm, or simplex method for local optimization. It uses a denser grid and a larger number of particles to improve the accuracy of concentration calculation. A regularization term is introduced into the objective function to constrain the variation of parameters in adjacent iterations to suppress solution oscillations caused by observation noise. The uncertainty quantification unit uses the local curvature of the objective function in the neighborhood of the optimal solution to estimate the posterior confidence interval of each parameter. It outputs a confidence ellipse at the leakage source location to visually represent the location uncertainty. When the area of ​​the confidence ellipse exceeds a preset threshold, it automatically pushes a prompt suggesting the deployment of temporary measurement points or manual on-site verification. When the nonlinearity of the objective function is too high, causing the curvature method confidence estimation to fail, Monte Carlo sampling is used to assist in evaluating the uncertainty.

9. The method for tracing the source of metallurgical gas leakage based on physical inversion and optimization solution as described in claim 7, characterized in that, The online verification and adaptive correction module includes an online verification unit, an adaptive correction unit, and a multi-leakage source identification unit. The online verification unit substitutes the optimal leakage source parameter set into the diffusion model to calculate the concentration prediction value for future periods, compares it with subsequent real-time observation values, sets the continuous monitoring window length, and calculates the normalized root mean square deviation between the predicted concentration and the measured concentration. When the deviation shows a monotonically increasing trend over multiple consecutive monitoring periods and exceeds a preset threshold, it is determined that there is a significant deviation in the source tracing result. When the adaptive correction unit determines that there is a deviation in the source tracing result, it uses the current optimal solution as the initial value and re-executes fine search optimization using newly added observation data to update the leakage source parameter estimation. The multi-leakage source identification unit sets a residual judgment threshold. When the optimal objective function value of a single leakage source model is continuously higher than the threshold, the multi-leakage source identification mode is activated, and the observation data is grouped or independently inverted in time segments to distinguish the locations and parameters of multiple leakage sources.

10. The method for tracing the source of metallurgical gas leakage based on physical inversion and optimization solution according to claim 7, characterized in that, The source tracing-driven emergency response module includes a hazard distribution generation unit, an evacuation route planning unit, and a disposal plan matching unit. The hazard distribution generation unit marks the coordinates of the leak source location on the plant's GIS map and quickly generates a rough distribution map of hazard diffusion centered on the leak source, combined with real-time wind direction and speed. The evacuation route planning unit automatically determines the emergency response level based on the leak source location, leakage rate, and substance type, and plans personnel evacuation routes radiating outward from the area surrounding the leak source. The evacuation routes automatically avoid high-risk diffusion areas centered on the leak source and incorporate constraints from the plant's fire lanes, work zones, and fixed post distribution, guiding personnel to safe areas in the crosswind or upwind direction. The disposal plan matching unit matches the optimal disposal measures from the contingency plan library based on the leaked substance type and the type of equipment where the leak source is located, generating a disposal plan that includes specific operations such as valve shut-off, sprinkler activation, and area warning. The disposal plan is bound to the leak source location to guide emergency personnel to quickly and accurately reach the target equipment for disposal.

Images