Amc pollution source localization method combining mobile detection and forward evolution model

Abstract

This invention provides an AMC (Air Pollution Source Location) method combining mobile detection and a forward evolution model, belonging to the field of environmental monitoring technology. The method includes the following steps: acquiring spatial parameters of the area to be monitored and the starting detection position of the mobile detection device, and planning the detection path of the mobile detection device by combining a grid normalization algorithm and uniform coverage constraints; based on the detection path of the mobile detection device, using the sensor-equipped mobile detection device to collect air environmental data of the area to be monitored, and recording the three-dimensional coordinates of the location of each air environmental data point to obtain an environment-location dataset; based on the spatial parameters of the area to be monitored, constructing a forward evolution model, combining the Haken model and the environment-location dataset for co-evolutionary analysis, and identifying the pollution source location by solving for the minimum point of the potential function. This invention, combining the forward evolution analysis of the Haken model and the solution of the minimum potential function, can accurately identify the pollution source location.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and more particularly to an AMC pollution source localization method that combines mobile detection with a forward evolution model. Background Technology

[0002] In semiconductor manufacturing, controlling gaseous molecular contaminants (AMCs) is crucial. AMCs adhere to the wafer surface, affecting the electrical performance, reliability, and yield of semiconductor devices, and can even lead to product scrap. Accurately and quickly locating AMC contamination sources within the semiconductor manufacturing workshop is a key step in effectively controlling contamination spread and ensuring semiconductor product quality.

[0003] However, semiconductor manufacturing workshops are complex environments, characterized by compact spatial layouts, dense equipment, and diverse processes. Furthermore, AMCs exhibit rapid diffusion and complex influencing factors, posing numerous challenges to traditional pollution source location methods.

[0004] Currently, common methods for locating pollution sources are mainly divided into fixed monitoring point methods and methods based on simple diffusion models. Fixed monitoring point methods involve pre-setting multiple fixed monitoring points in the semiconductor manufacturing workshop to collect air quality data, and then inferring the location of the pollution source based on data anomalies. However, this method has significant limitations: firstly, the number and location of fixed monitoring points are difficult to adjust flexibly according to the actual layout of the workshop, production processes, and pollution diffusion, easily leading to monitoring blind spots and an inability to comprehensively obtain pollution information; secondly, the construction and maintenance costs of fixed monitoring points are high, making them economically and temporally prohibitive for large or complex semiconductor manufacturing workshops.

[0005] There are currently no effective solutions to the problems in the relevant technologies. Summary of the Invention

[0006] In view of this, the present invention provides an AMC pollution source localization method that combines mobile detection and a forward evolution model to solve the aforementioned problems.

[0007] To solve the above problems, the specific technical solution adopted by the present invention is as follows:

[0008] A method for locating AMC pollution sources that combines mobile detection with a forward evolution model includes the following steps:

[0009] S1. Obtain the spatial parameters of the area to be detected and the starting detection position of the mobile detection device, and combine the grid normalization algorithm and uniform coverage constraint to plan the detection path of the mobile detection device;

[0010] S2. Based on the detection path of the mobile detection device, the mobile detection device equipped with sensors is used to collect air environment data of the area to be detected, and the three-dimensional coordinates of the location of each air environment data point are recorded to obtain the environment-location dataset.

[0011] S3. Based on the spatial parameters of the area to be detected, a forward evolution model is constructed. Combined with the Haken model and the environment-location dataset, a co-evolutionary analysis is performed. The location of the pollution source is identified by solving for the minimum point of the potential function.

[0012] Preferably, the steps of obtaining the spatial parameters of the area to be detected and the starting detection position of the moving detection device, and planning the detection path of the moving detection device in conjunction with the grid normalization algorithm and uniform coverage constraint, include the following steps:

[0013] S11. Obtain the three-dimensional spatial information of the area to be detected, and determine the unit mesh size by combining the size and mobility performance of the mobile detection device, and construct a three-dimensional spatial mesh model based on the unit mesh size;

[0014] S12. Based on the uniform coverage constraint, semantically label all grids in the three-dimensional spatial grid model and determine the path cost weight of each grid.

[0015] S13. Based on the normalized A* algorithm, combined with the starting detection position of the moving detection device, a normalized cost function for the moving detection device is constructed, and the optimal detection path for the moving detection device is generated using a parent-child grid search strategy.

[0016] Preferably, the step of constructing a normalized cost function for the motion detection device based on the normalized A* algorithm, combined with the starting detection position of the motion detection device, and generating the optimal detection path for the motion detection device using a parent-child grid search strategy includes the following steps:

[0017] S131. Based on the normalized A-star algorithm, the cost term of the mobile detection device is determined by utilizing the starting detection position of the mobile detection device, and a normalized cost function of the mobile detection device is constructed by combining heuristic estimation.

[0018] S132. Take the starting detection position of the mobile detection device as the starting point of the path search, and normalize its spatial coordinates to map them into three-dimensional grid coordinates.

[0019] S133. Based on the parent-child grid search strategy, perform path search and cost update, and expand the search by combining the minimum cost node selection mechanism, and generate the optimal detection path by backtracking the parent node chain.

[0020] Preferably, the step of constructing a forward evolution model based on the spatial parameters of the area to be detected, combining the Haken model and the environment-location dataset for co-evolutionary analysis, and identifying the location of the pollution source by solving for the minimum point of the potential function includes the following steps:

[0021] S31. Based on the spatial parameters of the area to be detected, the diffusion factors of gaseous pollutants are analyzed, and a forward evolution model of pollutant diffusion is constructed.

[0022] S32. Construct state variables using environment-location datasets and simulation results from a forward evolution model, and determine the evolutionary starting point;

[0023] S33. Based on the evolutionary starting point, establish a co-evolution equation between measured data and simulated data using the Haken model;

[0024] S34. By solving the minimum point of the potential function in the co-evolution equation, determine the pollution source location parameters that best match the measured and simulated data.

[0025] Preferably, the step of analyzing the diffusion factors of gaseous pollutants based on the spatial parameters of the area to be detected and constructing a forward evolution model of pollutant diffusion includes the following steps:

[0026] S311. Based on the spatial parameters of the area to be detected, identify and determine the diffusion factors that affect the diffusion of gaseous pollutants in the area to be detected;

[0027] S312. Using the grey comprehensive analysis method, analyze the causal relationship between various diffusion factors and the hierarchical structure among them, and identify the key factors affecting the diffusion process.

[0028] S313. Based on the causal relationships among various diffusion factors and the key factors affecting the diffusion process, and combined with dynamic Bayesian networks, a forward evolution model of pollutant diffusion is constructed.

[0029] Preferably, the step of using grey comprehensive analysis to analyze the causal relationships and hierarchical structure among various diffusion factors, and to identify key factors affecting the diffusion process, includes the following steps:

[0030] S3121. Obtain fuzzy evaluation data for each diffusion factor, and construct a gray direct influence matrix through gray interval transformation;

[0031] S3122. Normalize the gray direct influence matrix and calculate the comprehensive influence matrix through matrix operations.

[0032] S3123. Based on the comprehensive influence matrix and combined with the preset threshold, determine the causal relationship and hierarchical structure among the diffusion factors, and identify the key factors affecting the diffusion process.

[0033] Preferably, the step of constructing state variables using environment-location datasets and simulation results from a forward evolution model, and determining the evolutionary starting point, includes the following steps:

[0034] S321. Determine the physical boundary of the area to be detected based on the environment-location dataset, and delineate the three-dimensional spatial location range and release intensity range of the pollution source;

[0035] S322. Using the spatial grid division method, several candidate pollution sources with different locations are generated within the three-dimensional spatial location range;

[0036] S323. For each pollution candidate source, a forward evolution model is used to evolve it to obtain the simulated concentration field of each pollution candidate source. The simulated data sequence that is consistent with the spatiotemporal coordinates of the mobile detection path is extracted from the simulated concentration field of the pollution candidate source. The matching degree between the simulated data sequence and the measured concentration sequence in the environment-location dataset is calculated, and the matching degree is assigned as the initial value of the state variable of the pollution candidate source.

[0037] S324. Combine each pollution candidate source with its corresponding initial value of state variable to form an initial state set for co-evolution, and take the initial state set as the starting point of evolution.

[0038] Preferably, the step of establishing a co-evolutionary equation between measured and simulated data based on the evolutionary starting point using the Haken model includes the following steps:

[0039] S331. Based on the initial state set of the evolution starting point, determine the binary system of the Haken model and the order parameters and dependent variables of the binary system;

[0040] S332. Based on the order parameters and dependent variables of the twin system, construct the co-evolution equation between measured data and simulated data.

[0041] Preferably, determining the binary system of the Haken model and the order parameters and dependent variables of the binary system includes:

[0042] The measured concentration sequence in the environmental-location dataset collected by the mobile detection device is used as the first subsystem;

[0043] The simulated data sequence obtained by evolving the forward evolution model is used as the second subsystem;

[0044] The state variables of all pollution candidate sources are treated as order parameters;

[0045] The mean of all state variables is used as a subordinate variable.

[0046] Preferably, the step of determining the pollution source location parameters that optimally match the measured and simulated data by solving the potential function minimum point of the co-evolution equation includes the following steps:

[0047] S341. Based on the co-evolution equation, extract the evolution relationship of the order parameter, and based on the principle that the derivative of the function is equal to the negative value of the order parameter evolution equation, use integral operation to construct the potential function corresponding to the state variables of each pollution candidate source.

[0048] S342. Calculate and compare the potential function values ​​corresponding to all pollution candidate sources, and select the pollution candidate source with the smallest potential function value.

[0049] S343. Based on the pollution candidate source with the smallest potential function value, determine the location parameters of the pollution source.

[0050] The beneficial effects of this invention are as follows:

[0051] 1. This invention scientifically plans the detection path through a grid normalization algorithm, which enables efficient and uniform coverage of the area to be detected, avoiding data blind spots. It utilizes environmental-location datasets collected by mobile devices to provide high spatiotemporal resolution experimental support for the model. Combined with the forward evolution analysis of the Haken model and the solution of the minimum potential function, the location of pollution sources can be accurately identified.

[0052] 2. This invention acquires three-dimensional spatial information of the region and constructs a spatial grid model adapted to the performance of the equipment, which can ensure that the path planning is accurately matched with the actual environment. By using uniform coverage constraints and semantic tagging technology, the detection path can fully cover the area to be detected and avoid repetition, thereby improving data acquisition efficiency. By combining the normalized A* algorithm and the parent-child grid search strategy, the device maneuverability, path cost weight and heuristic estimation are incorporated into the cost function to achieve a fast solution for the globally optimal path.

[0053] 3. This invention constructs an accurate positive pollutant diffusion model by combining spatial parameters and multi-factor analysis. It uses the grey comprehensive analysis method to quantify the complex relationships between diffusion factors and identify key factors, ensuring the model's scientific reliability. Based on environmental and measured data, it constructs state variables and co-evolution starting points, and combines the Haken model to establish a dynamic matching mechanism between measured and simulated data, effectively capturing the co-evolutionary characteristics of pollution propagation. It achieves accurate location of pollution sources by solving for the minimum value of the potential function. The entire process integrates the advantages of data-driven and model-based deduction, significantly improving the accuracy, robustness, and interpretability of pollution source location in complex environments. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0055] Figure 1 This is a flowchart of an AMC pollution source localization method combining mobile detection and a forward evolution model according to an embodiment of the present invention. Detailed Implementation

[0056] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0057] According to embodiments of the present invention, an AMC pollution source localization method combining motion detection and a forward evolution model is provided.

[0058] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the AMC pollution source localization method combining motion detection and a forward evolution model according to an embodiment of the present invention includes the following steps:

[0059] S1. Obtain the spatial parameters of the area to be detected and the starting detection position of the mobile detection device, and combine the grid normalization algorithm and uniform coverage constraint to plan the detection path of the mobile detection device;

[0060] In a preferred embodiment, the steps of obtaining the spatial parameters of the area to be detected and the starting detection position of the moving detection device, and planning the detection path of the moving detection device in conjunction with the grid normalization algorithm and uniform coverage constraints, include the following steps:

[0061] S11. Obtain the three-dimensional spatial information of the area to be detected, and determine the unit mesh size by combining the size and mobility performance of the mobile detection device, and construct a three-dimensional spatial mesh model based on the unit mesh size;

[0062] It should be noted that the acquisition of three-dimensional spatial information can be achieved by using LiDAR to obtain the three-dimensional spatial information of the area to be detected (semiconductor manufacturing workshop), including three-dimensional spatial data containing detailed information such as the position, shape, and height of each object.

[0063] S12. Based on the uniform coverage constraint, semantically label all grids in the three-dimensional spatial grid model and determine the path cost weight of each grid.

[0064] Specifically, uniform coverage constraints enable the detection path to uniformly collect all grids in the 3D spatial grid model, while semantic labeling marks grid types based on functional or spatial characteristics, including passable areas, obstacle areas, equipment areas, etc.

[0065] The allocation of path cost weights needs to be combined with grid type and device status, such as power consumption and load, and the cost weights should be dynamically adjusted.

[0066] S13. Based on the normalized A* algorithm, combined with the starting detection position of the moving detection device, a normalized cost function for the moving detection device is constructed, and the optimal detection path for the moving detection device is generated using a parent-child grid search strategy.

[0067] As a preferred embodiment, the step of constructing a normalized cost function for the motion detection device based on the normalized A* algorithm, combined with the initial detection position of the motion detection device, and generating the optimal detection path for the motion detection device using a parent-child grid search strategy includes the following steps:

[0068] S131. Based on the normalized A-star algorithm, the cost term of the mobile detection device is determined by utilizing the starting detection position of the mobile detection device, and a normalized cost function of the mobile detection device is constructed by combining heuristic estimation.

[0069] It should be noted that the cost term needs to be determined based on the mobility performance of the mobile detection equipment, such as maximum speed, acceleration, and turning radius, to calculate the actual movement cost from the starting position to the current grid. For example, if there are no obstacles or equipment between grids, the cost is the Euclidean distance multiplied by the spatial weight to obtain the straight-line movement cost.

[0070] The heuristic function is used to estimate the remaining cost from the current grid to the target point.

[0071] S132. Take the starting detection position of the mobile detection device as the starting point of the path search, and normalize its spatial coordinates to map them into three-dimensional grid coordinates.

[0072] S133. Based on the parent-child grid search strategy, perform path search and cost update, and expand the search by combining the minimum cost node selection mechanism, and generate the optimal detection path by backtracking the parent node chain.

[0073] It should be noted that the path planning based on the parent-child grid search strategy can achieve efficient exploration and optimization by dynamically constructing a search tree. Specifically, this includes: taking the starting grid of the moving detection device as the root node, making it the parent node, and initializing its cost to 0; in each round of search, the algorithm selects the current minimum cost node from the priority queue as the expansion center, generating all passable sub-grids in its neighborhood; for each sub-grid, the actual cost of moving from the parent node to the sub-grid is calculated based on the device's maneuverability, and its total cost is updated using a heuristic function; if the sub-grid has been visited, its parent node pointer and cost are updated only when the cost of the new path is lower, to ensure that the search tree always retains the optimal path branch; by iteratively expanding the minimum cost node, the search range gradually converges towards the target grid until the target grid is selected as the node to be expanded; by tracing back the parent node chain from the target grid to the starting grid, the optimal detection path composed of continuous grid coordinates is constructed through reverse processing.

[0074] S2. Based on the detection path of the mobile detection device, the mobile detection device equipped with sensors is used to collect air environment data of the area to be detected, and the three-dimensional coordinates of the location of each air environment data point are recorded to obtain the environment-location dataset.

[0075] It should be noted that sampling positions are set at the center or boundary points of each grid according to the optimal detection path. The mobile detection device is equipped with various gas sensors, such as PM2.5 laser sensors. Electrochemical sensors, temperature and humidity probes, and anemometers are aligned via hardware synchronization signals or timestamps, so that all sensors are triggered to collect data simultaneously when the device reaches each sampling point.

[0076] S3. Based on the spatial parameters of the area to be detected, a forward evolution model is constructed. Combined with the Haken model and the environment-location dataset, a co-evolutionary analysis is performed. The location of the pollution source is identified by solving for the minimum point of the potential function.

[0077] As a preferred implementation, the step of constructing a forward evolution model based on the spatial parameters of the area to be detected, combining the Haken model and the environment-location dataset for co-evolutionary analysis, and identifying the location of the pollution source by solving for the minimum point of the potential function includes the following steps:

[0078] S31. Based on the spatial parameters of the area to be detected, the diffusion factors of gaseous pollutants are analyzed, and a forward evolution model of pollutant diffusion is constructed.

[0079] As a preferred embodiment, the step of analyzing the diffusion factors of gaseous pollutants based on the spatial parameters of the area to be detected and constructing a forward evolution model of pollutant diffusion includes the following steps:

[0080] S311. Based on the spatial parameters of the area to be detected, identify and determine the diffusion factors that affect the diffusion of gaseous pollutants in the area to be detected;

[0081] Specifically, in semiconductor manufacturing workshops, the diffusion factors affecting the diffusion of gaseous pollutants mainly include dynamic factors, thermal factors, pollutant source characteristics, and workshop layout. Dynamic factors refer to the design of the ventilation system (such as supply and exhaust air volumes), which directly affects the air exchange efficiency inside and outside the workshop. Insufficient ventilation can lead to pollutant accumulation in localized areas; excessive ventilation can result in decreased cleanliness or energy waste. Pollutant source characteristics include the large amount of gaseous molecular pollutants generated by processes such as photolithography, etching, and doping in semiconductor manufacturing. Thermal factors involve the heat generated by equipment operation in semiconductor workshops, which can lead to localized temperature increases, forming unstable stratification, enhancing turbulent activity, and promoting the vertical diffusion of pollutants. Workshop layout includes the pressure difference between the interior and exterior of the workshop or between areas with different cleanliness levels, and the arrangement of air supply and return vents to create stable laminar flow, etc.

[0082] S312. Using the grey comprehensive analysis method, analyze the causal relationship between various diffusion factors and the hierarchical structure among them, and identify the key factors affecting the diffusion process.

[0083] As a preferred embodiment, the step of using grey comprehensive analysis to analyze the causal relationships and hierarchical structure among various diffusion factors, and to identify key factors affecting the diffusion process, includes the following steps:

[0084] S3121. Obtain fuzzy evaluation data for each diffusion factor, and construct a gray direct influence matrix through gray interval transformation;

[0085] Specifically, obtaining fuzzy assessment data for each diffusion factor can be achieved by having 5-10 domain experts score the direct influence strength between each factor. These expert scores are then converted into gray-zone numbers. ,in, a The minimum possible value, such as the lower limit of expert ratings; b This represents the maximum possible value, such as the upper limit of expert ratings.

[0086] S3122. Normalize the gray direct influence matrix and calculate the comprehensive influence matrix through matrix operations.

[0087] It should be noted that the elements in the gray direct influence matrix are gray interval numbers, representing the range of the direct influence intensity between factors. To unify the units and simplify the calculation, the matrix needs to be normalized: each gray interval is divided by its global maximum upper bound (i.e., the maximum value of the upper bounds of all intervals in the matrix) to obtain the normalized intervals, forming a normalized matrix.

[0088] Among these, the comprehensive influence matrix is ​​calculated based on the normalized matrix. This matrix needs to reflect both the direct and indirect influences between factors (such as factors...). i Through factors k Factors j (Indirect effects). Specifically, matrix operations can be performed according to the matrix operation principles of the DEMATEL method. The comprehensive influence matrix is ​​solved by the sum of infinite series of the normalized matrix, as shown in the formula:

[0089] U = O + O 2+ O 3+...+ O m ( m →∞).

[0090] When matrix O When the spectral radius is less than 1, this infinite series converges and can be simplified to... U = O ( IO )-1, where, I express n × n identity matrix, ( IO )-1 means ( IO The inverse matrix of ) is obtained. Through matrix multiplication and inversion operations, the final comprehensive influence matrix is ​​obtained. U =[ u ij ] n × n In the matrix u ij The larger the value, the more significant the factor. i Factors j The stronger the combined direct and indirect impact, the greater the overall effect.

[0091] S3123. Based on the comprehensive influence matrix and combined with the preset threshold, determine the causal relationship and hierarchical structure among the diffusion factors, and identify the key factors affecting the diffusion process.

[0092] It should be noted that the comprehensive impact matrix U elements u ij Quantified factors i Factors j The total influence strength (normalized to [0,1]), while the budget threshold can be calculated by taking the mean of all elements. m with standard deviation s And through the threshold calculation formula i = m + ks ( k(The confidence coefficient is typically set between 1.5 and 2.0) is used to calculate the preset threshold. i .

[0093] The rules for determining causality are as follows:

[0094] like u ij ≥ i Then the determining factors i Factors j A significant causal relationship exists, denoted as i → j ;

[0095] like u ij < i If the influence is negligible, it is not included in the causal network.

[0096] Furthermore, after threshold filtering, the remaining significant influence relationships can be represented as a directed graph. G =( V , E ), where the vertex set V For the set of diffusion factors, edge set E ={( i , j )∣ t ij ≥ i Based on this graph, a hierarchical decomposition is performed, specifically including: calculating the in-degree (number of edges pointing to that vertex) and out-degree (number of edges originating from that vertex) of all vertices; assigning vertices with an in-degree of 0 to level 1 (the root node), removing these vertices and their out-edges, and updating the in-degree of the remaining vertices; repeating this process until all vertices are assigned a level.

[0097] Furthermore, key factors are the core nodes influencing propagation within the causal network. Their identification requires a comprehensive consideration of the strength of both direct and indirect influences. By calculating degree centrality and betweenness centrality, vertices ranking in the top 30% for both indicators are selected as key factors. These factors are not only core nodes with a wide direct impact on the causal network but also intermediary hubs connecting factors at different levels, exerting a dominant influence on the entire diffusion process. Degree centrality refers to the sum of a vertex's in-degree and out-degree, while betweenness centrality refers to the number of times a vertex lies on the shortest path between any two other vertices.

[0098] S313. Based on the causal relationships among various diffusion factors and the key factors affecting the diffusion process, and combined with dynamic Bayesian networks, a forward evolution model of pollutant diffusion is constructed.

[0099] Specifically, when constructing a forward evolution model for pollutant diffusion, the previously identified causal relationships are directly transformed into a DBN topology. The DBN topology consists of nodes and directed edges; nodes correspond to diffusion factors, and directed edges correspond to causal directions. Simultaneously, based on a hierarchical structure, bottom-level basic factors are set as parent nodes, and top-level outcome factors are set as child nodes, forming a static topology framework. A time dimension is introduced, expanding each factor node into a multi-time-slice node, connecting nodes of the same factor in different time slices through directed edges, thus completing the dynamic topology construction. Secondly, for static nodes, prior probabilities are determined based on statistical analysis of spatial parameters of the area to be detected; for dynamic nodes, prior probability distributions are fitted using historical meteorological data and diffusion experiment data; and for the conditional probability table between nodes, it is set using the strength of causal relationships and physical diffusion laws.

[0100] Simultaneously, edge probabilities are calibrated using expert experience to ensure parameters conform to actual diffusion scenarios. Real-time monitoring data of key factors are used as observational evidence for the DBN, and forward inference is used to predict the spatiotemporal distribution of pollutant concentrations in different time slices. When new environmental-location datasets are acquired, backward inference is used to correct node probability parameters, dynamically optimizing the model's simulation accuracy, ultimately forming a forward model that can evolve in real time and closely matches actual diffusion patterns.

[0101] S32. Construct state variables using environment-location datasets and simulation results from a forward evolution model, and determine the evolutionary starting point;

[0102] In a preferred embodiment, the step of constructing state variables using environment-location datasets and simulation results from a forward evolution model, and determining the evolutionary starting point, includes the following steps:

[0103] S321. Determine the physical boundary of the area to be detected based on the environment-location dataset, and delineate the three-dimensional spatial location range and release intensity range of the pollution source;

[0104] Specifically, when determining the physical boundaries of the area to be detected based on the environment-location dataset, the layout of the workshop's architectural structure, such as the location of walls, doors, and windows, as well as the division of different functional areas, such as the lithography area, etching area, and packaging and testing area, are considered to determine the isolation and connectivity between these areas. This allows for the precise definition of the physical range of the area to be detected in both the horizontal and vertical directions. Regarding the delineation of the three-dimensional spatial location of pollution sources, the locations of equipment that may generate gaseous pollutants are analyzed in conjunction with various semiconductor manufacturing processes, such as the use of photoresist in the lithography process and the emission of chemical gases in the etching process. This includes the placement height of the equipment and its distribution coordinates within the workshop. Simultaneously, potential pollution sources from personnel activity areas are considered, such as pollutants carried by personnel or particles generated during operations. This allows for the determination of the location range of pollution sources in the length, width, and height dimensions. The determination of the release intensity range is based on the equipment's operating parameters, such as gas flow rate and power, as well as the characteristics of the process, such as the intensity of the reaction and reaction time. It also takes into account the changes in pollutant concentration in historical monitoring data. Through professional simulation calculations and data analysis methods, the upper and lower limits of pollutant release intensity from different pollution sources under different operating conditions are determined, providing an accurate basis for subsequent pollution prevention and control and detection work.

[0105] S322. Using the spatial grid division method, several candidate pollution sources with different locations are generated within the three-dimensional spatial location range;

[0106] It should be noted that each grid node corresponds to a candidate pollution source, and its coordinates are the center point of the grid.

[0107] S323. For each pollution candidate source, a forward evolution model is used to evolve it to obtain the simulated concentration field of each pollution candidate source. The simulated data sequence that is consistent with the spatiotemporal coordinates of the mobile detection path is extracted from the simulated concentration field of the pollution candidate source. The matching degree between the simulated data sequence and the measured concentration sequence in the environment-location dataset is calculated, and the matching degree is assigned as the initial value of the state variable of the pollution candidate source.

[0108] It should be noted that for each preset pollution candidate source, dynamic simulation is performed through a forward evolution model. The input parameters include the three-dimensional coordinates of the pollution candidate source, the preset release intensity range, and the real-time environmental conditions provided in the environment-location dataset, thereby generating the spatiotemporal concentration field distribution of the pollution candidate source in the area to be detected. Subsequently, based on the spatiotemporal coordinates of the moving detection path, a spatiotemporal point data sequence that perfectly matches the measured path is accurately extracted from the simulated concentration field, ensuring that the simulated sequence and the measured sequence correspond one-to-one in the spatiotemporal dimension. The similarity between the simulated sequence and the measured concentration sequence is quantified by statistical indicators, and the calculated matching degree value will be used as the initial state variable of the pollution candidate source.

[0109] S324. Combine each pollution candidate source with its corresponding initial value of state variable to form an initial state set for co-evolution, and take the initial state set as the starting point of evolution.

[0110] It should be noted that each generated candidate pollution source is structurally integrated with the calculated initial values ​​of the state variables to construct a tuple of pollution candidate source-matching degree, forming an initial state set containing information on all pollution candidate sources.

[0111] S33. Based on the evolutionary starting point, establish a co-evolution equation between measured data and simulated data using the Haken model;

[0112] As a preferred implementation, the step of establishing a co-evolutionary equation between measured and simulated data based on the evolutionary starting point using the Haken model includes the following steps:

[0113] S331. Based on the initial state set of the evolution starting point, determine the binary system of the Haken model and the order parameters and dependent variables of the binary system;

[0114] It should be noted that determining the binary system, order parameters, and dependent variables of the Haken model includes:

[0115] The measured concentration sequence in the environmental-location dataset collected by the mobile detection device is used as the first subsystem to reflect the spatiotemporal dynamics of real pollution diffusion.

[0116] The simulated data sequence obtained by evolving the forward evolution model is used as the second subsystem, reflecting the theoretical pollution diffusion pattern.

[0117] The state variables of all pollution candidate sources are used as order parameters, which are global variables that characterize the cooperative behavior of the two subsystems and drive the evolution of the system from disorder to order.

[0118] The mean of all state variables is used as a subordinate variable. The mean of the state variables can quantify the overall degree of matching, and as a subordinate variable, it reflects the level of coordination between the measured and simulated results.

[0119] S332. Based on the order parameters and dependent variables of the twin system, construct the co-evolution equation between measured data and simulated data.

[0120] It should be noted that the basic equations of the Haken model are:

[0121] ;

[0122] ;

[0123] In the formula, q i This represents the order parameter, i.e., the degree of matching with the pollution candidate sources.s j This represents the dependent variable, i.e., the mean matching degree. c i The attenuation coefficient represents the order parameter (reflecting the stability of the pollution candidate source). f i , g j This represents a nonlinear coupling function that describes the interaction between measured and simulated data. x i ( t ) represents random perturbation (simulated detection error).

[0124] The co-evolution equations between measured and simulated data include: order parameter evolution equations and dependent variable evolution equations;

[0125] The formula for calculating the order parameter evolution equation is:

[0126] ;

[0127] In the formula, α Indicates the matching degree decay rate. β represents the synergy coefficient, and Similarity represents the similarity function.

[0128] The formula for calculating the evolution equation of dependent variables is:

[0129] ;

[0130] In the formula, M i Indicates the first i The degree of matching of each pollution candidate source N Indicates the number of candidate pollution sources. M avg This represents the mean match rate.

[0131] S34. By solving the minimum point of the potential function in the co-evolution equation, determine the pollution source location parameters that best match the measured and simulated data.

[0132] In a preferred embodiment, determining the pollution source location parameters that optimally match the measured and simulated data by solving the minimum point of the potential function in the co-evolution equation includes the following steps:

[0133] S341. Based on the co-evolution equation, extract the evolution relationship of the order parameter, and based on the principle that the derivative of the function is equal to the negative value of the order parameter evolution equation, use integral operation to construct the potential function corresponding to the state variables of each pollution candidate source.

[0134] It should be noted that when constructing the potential function of pollution candidate sources based on the co-evolution equation, the dynamic evolution relationship of the order parameter is extracted from the co-evolution equation. The evolution of the order parameter is usually determined by the internal synergistic effect and external driving force of the system. Its mathematical form is a differential equation containing a decay term (reflecting the system's memory) and a synergistic term (reflecting data matching drive). According to the physical principle of potential function construction, when the system reaches a steady state, the time derivative of the order parameter should be equal to the negative gradient of the potential function. That is, by making the derivative term of the order parameter evolution equation equal to the negative gradient of the potential function, the relationship between the differential equation and the potential function can be established. Then, the relationship is solved in reverse by using integral operation, transforming the non-potential function terms in the order parameter evolution equation into components of the potential function through integration, while retaining the quadratic form of the decay term, and finally obtaining the potential function expression corresponding to the state variables of each pollution candidate source.

[0135] S342. Calculate and compare the potential function values ​​corresponding to all pollution candidate sources, and select the pollution candidate source with the smallest potential function value.

[0136] It should be noted that the potential function model constructed based on the co-evolution equation transforms the dynamic matching process of simulated data and measured data of each pollution candidate source into a quantitative potential energy value. The smaller the potential function value, the more stable the co-evolution of the system, that is, the higher the degree of fit between the pollution candidate source and the actual pollution source.

[0137] Specifically, the potential function values ​​of all pollution candidate sources are obtained through numerical calculation, and the dimensional differences are eliminated by normalization. The pollution candidate source with the smallest potential function value is selected as the optimal solution. Its physical meaning is that the pollution source can make the simulated and measured data reach the lowest energy stable state in the long-term evolution, thus providing the theoretically optimal location result for pollution source tracing.

[0138] S343. Based on the pollution candidate source with the smallest potential function value, determine the location parameters of the pollution source.

[0139] Specifically, the minimum potential function of the pollution candidate source reflects the optimal coordination state between its simulated and measured data in the spatiotemporal dimension. This state is directly related to parameters such as the spatial coordinates, release intensity, and temporal characteristics of the pollution source. By substituting the spatial diffusion model or data matching function into the co-evolution equation, key parameters such as the center location and diffusion range of the pollution source can be calculated. Combined with spatial information systems, visualization calibration can be performed, ultimately achieving precise location and parameterized description of the pollution source.

[0140] In summary, by utilizing the above-mentioned technical solutions of this invention, the present invention scientifically plans the detection path through a grid normalization algorithm, enabling efficient and uniform coverage of the detection area and avoiding data blind spots. It leverages environmental-location datasets collected by mobile devices to provide high spatiotemporal resolution experimental support for the model. Combined with the forward evolution analysis and potential function minimization of the Haken model, the location of pollution sources can be accurately identified. This invention, by acquiring three-dimensional spatial information of the region and constructing a spatial grid model adapted to device performance, ensures accurate matching between path planning and the actual environment. Utilizing uniform coverage constraints and semantic tagging technology, the detection path fully covers the detection area and avoids duplication, improving data acquisition efficiency. By combining the normalized A* algorithm with a parent-child grid search strategy, it integrates device mobility performance, path cost weights, and heuristic estimation into the cost function, achieving rapid solution for the globally optimal path. This invention constructs an accurate positive pollutant diffusion model by combining spatial parameters and multi-factor analysis. It uses the grey comprehensive analysis method to quantify the complex relationships between diffusion factors and identify key factors, ensuring the model's scientific reliability. Based on environmental and measured data, it constructs state variables and co-evolution starting points, and combines the Haken model to establish a dynamic matching mechanism between measured and simulated data, effectively capturing the co-evolutionary characteristics of pollution propagation. By solving for the minimum value of the potential function, it achieves accurate location of pollution sources. The entire process integrates the advantages of data-driven and model-based deduction, significantly improving the accuracy, robustness, and interpretability of pollution source location in complex environments.

[0141] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0142] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for locating AMC pollution sources by combining mobile detection and a forward evolution model, characterized in that, The method includes the following steps: S1. Obtain the three-dimensional spatial information of the area to be detected, and determine the unit grid size by combining the size and mobility performance of the mobile detection device. Construct a three-dimensional spatial grid model based on the unit grid size. Based on the uniform coverage constraint, semantically label all grids in the three-dimensional spatial grid model and determine the path cost weight of each grid. Based on the normalized A* algorithm and combined with the starting detection position of the mobile detection device, construct the normalized cost function of the mobile detection device, and generate the optimal detection path of the mobile detection device using the parent-child grid search strategy. Among them, the uniform coverage constraint enables the detection path to uniformly collect all the meshes in the three-dimensional spatial mesh model; S2. Based on the detection path of the mobile detection device, the mobile detection device equipped with sensors is used to collect air environment data of the area to be detected, and the three-dimensional coordinates of the location of each air environment data point are recorded to obtain the environment-location dataset. S3. Based on the spatial parameters of the area to be detected, identify and determine the diffusion factors affecting the diffusion of gaseous pollutants in the area to be detected; obtain fuzzy assessment data of each diffusion factor, and construct a gray direct influence matrix through gray interval transformation; normalize the gray direct influence matrix, and calculate the comprehensive influence matrix through matrix operations; based on the comprehensive influence matrix and combined with preset thresholds, determine the causal relationship and hierarchical structure among each diffusion factor, and identify the key factors affecting the diffusion process; based on the causal relationship among each diffusion factor and the key factors affecting the diffusion process, combine with a dynamic Bayesian network to construct a forward evolution model of pollutant diffusion; combine with the Haken model and environment-location dataset to conduct co-evolution analysis, and identify the location of pollution sources by solving for the minimum point of the potential function.

2. The AMC pollution source localization method combining mobile detection and a forward evolution model according to claim 1, characterized in that, The process of constructing a normalized cost function for the motion detection device based on the normalized A* algorithm, combined with the initial detection position of the motion detection device, and generating the optimal detection path for the motion detection device using a parent-child grid search strategy includes the following steps: S131. Based on the normalized A-star algorithm, the cost term of the mobile detection device is determined by utilizing the starting detection position of the mobile detection device, and a normalized cost function of the mobile detection device is constructed by combining heuristic estimation. S132. Take the starting detection position of the mobile detection device as the starting point of the path search, and normalize its spatial coordinates to map them into three-dimensional grid coordinates. S133. Based on the parent-child grid search strategy, perform path search and cost update, and expand the search by combining the minimum cost node selection mechanism, and generate the optimal detection path by backtracking the parent node chain.

3. The AMC pollution source localization method combining mobile detection and a forward evolution model according to claim 1, characterized in that, The method of combining the Haken model and environment-location dataset for co-evolutionary analysis, and identifying pollution source locations by solving for the minimum point of the potential function, includes the following steps: State variables are constructed using environment-location datasets and simulation results from a forward evolution model, and the evolutionary starting point is determined. Based on the evolutionary starting point, a co-evolutionary equation between measured data and simulated data is established using the Haken model; By solving for the minimum point of the potential function in the co-evolution equation, the pollution source location parameters that best match the measured and simulated data are determined.

4. The AMC pollution source localization method combining mobile detection and a forward evolution model according to claim 3, characterized in that, The process of constructing state variables and determining the evolutionary starting point using environment-location datasets and simulation results from a forward evolution model includes the following steps: The physical boundaries of the area to be detected are determined based on the environment-location dataset, and the three-dimensional spatial location range and release intensity range of the pollution source are delineated. Using a spatial grid partitioning method, several candidate pollution sources with different locations are generated within a three-dimensional spatial range; For each pollution candidate source, a forward evolution model is used to evolve it to obtain a simulated concentration field for each pollution candidate source. A simulated data sequence consistent with the spatiotemporal coordinates of the mobile detection path is extracted from the simulated concentration field of the pollution candidate source. The matching degree between the simulated data sequence and the measured concentration sequence in the environment-location dataset is calculated, and the matching degree is assigned as the initial value of the state variable of the pollution candidate source. Each pollution candidate source is combined with its corresponding initial state variable to form an initial state set for co-evolution, and this initial state set is used as the starting point for evolution.

5. The AMC pollution source localization method combining mobile detection and a forward evolution model according to claim 4, characterized in that, The process of establishing a co-evolutionary equation between measured and simulated data based on the evolutionary starting point and using the Haken model includes the following steps: Based on the initial state set at the starting point of evolution, the binary system of the Haken model and the order parameters and dependent variables of the binary system are determined. Based on the order parameters and dependent variables of the twin system, a co-evolution equation between measured and simulated data is constructed.

6. The AMC pollution source localization method combining mobile detection and a forward evolution model according to claim 5, characterized in that, The determination of the binary system in the Haken model, as well as the order parameters and dependent variables of the binary system, includes: The measured concentration sequence in the environmental-location dataset collected by the mobile detection device is used as the first subsystem; The simulated data sequence obtained by evolving the forward evolution model is used as the second subsystem; The state variables of all pollution candidate sources are treated as order parameters; The mean of all state variables is used as a subordinate variable.

7. The AMC pollution source localization method combining mobile detection and a forward evolution model according to claim 3, characterized in that, The process of determining the pollution source location parameters that optimally match measured and simulated data by solving the potential function minimum point of the co-evolution equation includes the following steps: Based on the co-evolution equation, the evolution relationship of the order parameter is extracted, and based on the principle that the derivative of the function is equal to the negative value of the order parameter evolution equation, the potential function corresponding to the state variables of each pollution candidate source is constructed by integral operation. Calculate and compare the potential function values ​​corresponding to all pollution candidate sources, and select the pollution candidate source with the smallest potential function value; Based on the pollution candidate source with the smallest potential function value, the location parameters of the pollution source are determined.

Images