AMC sampling point distribution optimization method based on multi-pollution source singular point fusion

Abstract

The application discloses an AMC sampling point distribution optimization method based on multi-pollution source singular point fusion, relates to the field of molecular pollution sampling, and comprises the following steps: extracting the respiratory volume and skin secretion volume generated by an operator as a molecular pollution load; coupling molecular pollution concentration data in a historical operation process in a monitoring area and the corresponding number of operators to obtain a causal relationship between the molecular pollution load and the molecular pollution concentration; obtaining the number and distribution positions of operators in the monitoring area in a target monitoring period, combining the number of operators with the causal relationship to predict the molecular pollution concentration in the monitoring area, and determining a multi-pollution source singular point; generating an initial distribution diagram of monitoring sampling points, and optimizing the initial distribution diagram according to the application cost under the condition of the initial distribution diagram to determine the final monitoring sampling points. The application can not only reduce monitoring cost, but also improve the efficiency of the sampling point layout and ensure that the key areas of the monitoring area are sufficiently covered.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of molecular contamination sampling, in particular to an AMC sampling point distribution optimization method based on multi-pollution source singular point fusion. BACKGROUND

[0002] AMC refers to air-borne molecular pollutants, including organic and inorganic pollutants in clean room air, existing in the form of gas, vapor and air dust, such as acid, alkali, polymer additives, organic metal compounds, etc., and the AMC pollutants are usually the key pollution sources in the high-precision manufacturing industry, and the products of the high-precision manufacturing industry are very sensitive to small pollutants in the environment, and any small molecular pollutants (such as acid, alkali, polymer additives, metal compounds, etc.) may affect the quality and performance of the products, and in order to obtain the concentration of AMC in the air, a sampling monitor is usually used to monitor the air.

[0003] The purpose of AMC sampling is to monitor and control the air-borne molecular pollutants in the controlled environment such as clean room to ensure that the environmental quality meets the strict requirements of the specific industry, but the existing AMC monitoring sampling points are usually designed according to the layout form of the monitoring area in the layout process, and the monitoring application cost in the monitoring process is ignored, so that the maximum monitoring efficiency cannot be realized within the narrow budget limit, the operation cost is increased, and the rationality of resource allocation is reduced. SUMMARY

[0004] In order to solve the above problems, the present application provides an AMC sampling point distribution optimization method based on multi-pollution source singular point fusion, which realizes the purpose of reducing the monitoring cost while improving the efficiency of the sampling point layout.

[0005] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0006] The present application provides an AMC sampling point distribution optimization method based on multi-pollution source singular point fusion, comprising:

[0007] Simulating the working behavior of the operator in the monitoring area, and extracting the respiratory volume and skin secretion volume of the operator in the target time period according to the working behavior simulation result, as the molecular pollution load;

[0008] Extracting the molecular pollutant concentration data in the historical operation process in the monitoring area, and coupling the corresponding number of operators, to obtain the causal relationship between the molecular pollution load and the molecular pollutant concentration based on the coupling result;

[0009] Obtaining the number and distribution position of the operators in the monitoring area in the target monitoring period, and combining the number of operators with the causal relationship to predict the molecular pollutant concentration in the monitoring area, and determining the multi-pollution source singular point;

[0010] Generate an initial distribution map of monitoring sampling points based on the singular points of multiple pollution sources and the distributed positions of operators, and optimize the initial distribution map to determine the final monitoring sampling points according to the application cost under the condition of the initial distribution map.

[0011] Preferably, the working behavior of the operator in the monitoring area is simulated, and the respiratory volume and skin secretion volume generated by the operator within the target time period are extracted as the molecular pollution load according to the simulation result of the working behavior, including:

[0012] Generate the working task of the operator based on the equipment working content of the monitoring area, decompose the working task into behavior action nodes, and set action running parameters for the behavior action nodes to obtain execution condition nodes;

[0013] Create a blank behavior tree, add the execution condition nodes to the blank behavior tree to generate a behavior rule engine, associate the behavior rule engine with the entity operator, and run the blank behavior tree to obtain the working behavior of the operator;

[0014] Obtain the respiratory frequency and sweating degree of the operator under the working behavior condition, construct a behavior-physiology coupling model, predict the respiratory volume and skin secretion volume generated by the operator within the target time period, and obtain the molecular pollution load.

[0015] Preferably, the working behavior of the operator in the monitoring area is simulated, and the respiratory volume and skin secretion volume generated by the operator within the target time period are extracted as the molecular pollution load according to the simulation result of the working behavior, including:

[0016] Start a behavior tree creation wizard based on the autonomous behavior rule editor, generate a blank behavior tree framework containing several root nodes and sub-node containers, and define behavior logic for the root nodes and sub-nodes using the rule editor;

[0017] Obtain the blank behavior tree according to the behavior logic definition result, add the execution condition nodes to the blank behavior tree nodes to extract operation instruction information, and generate a behavior rule engine based on the extraction result;

[0018] Associate the behavior rule engine with the entity operator simulation model, and send the operation instruction information to the entity operator simulation model to determine the behavior path of the entity operator in the virtual monitoring area;

[0019] Run the blank behavior tree structure in the simulation environment according to the behavior path of the entity operator, and obtain the working behavior of the operator in the virtual monitoring area based on the running result.

[0020] Preferably, the respiratory frequency and the sweating degree of the operator under the working behavior condition are acquired, a behavior-physiology coupling model is constructed, the respiratory volume and the skin secretion volume generated by the operator within a target time length are predicted, and the molecular contamination load is obtained, including:

[0021] The respiratory frequency and the sweating degree of the operator are extracted based on the working behavior in the virtual monitoring area, and the respiratory frequency and the sweating degree are combined to generate a bivariate observation sequence;

[0022] The bivariate observation sequence is subjected to cluster analysis by iteratively optimizing the cluster center with the Euclidean distance as the similarity measure, and the physiological behavior mode categories of the bivariate observation sequence are divided according to the cluster analysis results;

[0023] The behavior coupling model is established based on the physiological behavior mode categories by using the Bayesian prior distribution technology, and the posterior distribution of the behavior coupling model is sampled by using the Markov Chain Monte Carlo technology;

[0024] The respiratory frequency and the sweating degree of the operator within the target time length are estimated according to the sampling results, the respiratory frequency and the sweating degree are converted into the total volume of respiratory gas and the amount of skin sweat secretion by integrating the changes, the time-varying physiological output sequence is generated, and the molecular contamination load is obtained.

[0025] Preferably, the molecular contamination concentration data in the historical operation process in the monitoring area and the corresponding number of operators are coupled, and the causal relationship between the molecular contamination load and the molecular contamination concentration is obtained based on the coupling results, including:

[0026] The historical contamination concentration data and the number of operators in the monitoring area are collected, the historical molecular contamination concentration data and the number of operators are divided into several time periods, the fluctuation characteristics of the historical molecular contamination concentration data and the change mode of the operators in each time period are judged, and a time series data set is generated;

[0027] Based on the time series data set and the molecular contamination load, a neural network is used for time series modeling processing to capture the long-term dependence relationship between the historical molecular contamination concentration change and the molecular contamination load, and a causal network is generated;

[0028] The causal network is subjected to sensitivity analysis to evaluate the contribution degree of the number of operators in the causal network, the causal network is optimized according to the contribution degree, and the potential causal relationship between the time series data set and the molecular contamination load is captured.

[0029] Preferably, an initial distribution map of the monitoring sampling points is generated based on the singular points of the multiple pollution sources and the distribution positions of the operators, and the initial distribution map is optimized to determine the final monitoring sampling points according to the application cost under the condition of the initial distribution map, including:

[0030] Determine the upper limit value of the concentration of the molecular pollutants in the monitoring area based on the singular points of multiple pollution sources, and determine the maximum number of monitoring sampling points according to the upper limit value and the activity intensity of the distribution position of the operator;

[0031] According to the maximum number of monitoring sampling points and the layout state of the monitoring area, an axis-aligned bounding box is constructed by using the axis-aligned bounding box and directional bounding box technology, and an initial monitoring sampling point distribution map is generated based on the bounding box;

[0032] The purchase price of the monitoring sampling instrument is obtained, and the application cost of the initial monitoring sampling point distribution map is calculated according to the purchase price, and a minimum objective function is established based on the application cost, and the initial distribution map is optimized to determine the final monitoring sampling point.

[0033] Preferably, according to the maximum number of monitoring sampling points and the layout state of the monitoring area, an axis-aligned bounding box is constructed by using the axis-aligned bounding box and directional bounding box technology, and an initial monitoring sampling point distribution map is generated based on the bounding box;

[0034] The maximum number of monitoring sampling points and the layout state of the monitoring area are obtained, and the axis-aligned bounding box and directional bounding box construction technology are used to determine the optimal coordinate axis direction in combination with the layout state of the monitoring area;

[0035] Based on the optimal coordinate axis direction, a bounding box completely covering the monitoring area is constructed, and the monitoring sampling points are projected onto the nearest plane of the bounding box according to the layout state of the monitoring area, and the positions of the monitoring sampling points are output;

[0036] According to the position of the monitoring sampling point, the distance between the monitoring sampling point and the entity in the monitoring area is judged, and the monitoring sampling point with a distance higher than the distance threshold is removed, and the removed monitoring sampling point is re-projected;

[0037] Until the difference between the position of the monitoring sampling point and the distance of the entity in the monitoring area is less than or equal to the distance position, the initial monitoring sampling point distribution map is obtained.

[0038] The beneficial effects of the present application are:

[0039] 1、The present application establishes the causal relationship between pollution load and pollutant concentration, accurately reveals the dynamic change law between the two, provides a more scientific basis for future monitoring and prediction, and more accurately predicts the pollutant concentration in the target monitoring period after combining the number and distribution position of the operators in the monitoring area, accurately identifies the singular points of multiple pollution sources, lays a foundation for optimizing the sampling point layout, and optimizes the initial monitoring sampling point distribution map, which not only reduces the monitoring cost, but also improves the efficiency of the sampling point layout, ensures that the key areas of the monitoring area are covered enough, and avoids the waste of redundant sampling points.

[0040] 2、The application can accurately reflect the relationship between the pollutant concentration and the operator distribution, dynamically adjust the number of sampling points according to the pollution load of different regions, ensure that important regions are adequately monitored, thereby avoiding excessive monitoring of invalid regions, while ensuring that the monitoring of key pollution sources is not missed, thereby optimizing the monitoring efficiency, and evaluating the actual input cost of each sampling point, and establishing a minimum objective function on the basis of the actual input cost, further optimizing the initial sampling point layout, ensuring that the final sampling point layout can meet the demand for efficient monitoring, and also realize the maximum monitoring efficiency within the budget limit, which can effectively reduce the operating cost, improve the rationality of resource allocation, make the environmental monitoring more sustainable and efficient. BRIEF DESCRIPTION OF DRAWINGS

[0041] The accompanying drawings, which form a part of the present description, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of the application, and their

[0042] Figure 1 is a flow chart of the AMC sampling point distribution optimization method based on the multi-pollution source singular point fusion according to the embodiment of the application. DETAILED DESCRIPTION

[0043] The application will be further described below in conjunction with the drawings and embodiments.

[0044] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0045] It should be noted that the terms used herein are only intended to describe specific embodiments and are not intended to limit the exemplary embodiments according to the application. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form, and in addition, it should be understood that the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0046] The embodiments in the application and the features in the embodiments can be combined with each other without conflict.

[0047] Please refer to Figure 1The application provides an AMC sampling point distribution optimization method based on multi-pollution source singularity fusion, comprising the following steps:

[0048] In step S1, the working behavior of the operator in the monitoring area is simulated, and the respiratory volume and skin secretion volume of the operator in a target time period are extracted as molecular pollution load according to the working behavior simulation result.

[0049] In one embodiment, the working behavior of the operator in the monitoring area is simulated, and the respiratory volume and skin secretion volume of the operator in a target time period are extracted as molecular pollution load according to the working behavior simulation result, comprising the following steps: generating the working task of the operator based on the equipment working content of the monitoring area, decomposing the working task into behavior action nodes, setting action running parameters for the behavior action nodes to obtain execution condition nodes; creating a blank behavior tree, adding the execution condition nodes to the blank behavior tree to generate a behavior rule engine, and associating the behavior rule engine with the entity operator, running the blank behavior tree to obtain the working behavior of the operator; obtaining the respiratory frequency and sweating degree of the operator under the working behavior condition, constructing a behavior physiology coupling model, predicting the respiratory volume and skin secretion volume of the operator in a target time period, and obtaining the molecular pollution load.

[0050] In one embodiment, the working behavior of the operator in the monitoring area is simulated, and the respiratory volume and skin secretion volume of the operator in a target time period are extracted as molecular pollution load according to the working behavior simulation result, comprising the following steps: generating the working task of the operator based on the equipment working content of the monitoring area, decomposing the working task into behavior action nodes, setting action running parameters for the behavior action nodes to obtain execution condition nodes; creating a blank behavior tree, adding the execution condition nodes to the blank behavior tree to generate a behavior rule engine, and associating the behavior rule engine with the entity operator, running the blank behavior tree to obtain the working behavior of the operator; obtaining the respiratory frequency and sweating degree of the operator under the working behavior condition, constructing a behavior physiology coupling model, predicting the respiratory volume and skin secretion volume of the operator in a target time period, and obtaining the molecular pollution load.

[0051] In one embodiment, the respiratory frequency and sweating degree of the operator under the working behavior condition are acquired, a behavior physiological coupling model is constructed, the respiratory volume and skin secretion volume generated by the operator within a target time period are predicted, and the molecular contamination load is obtained, including: based on the working behavior in the virtual monitoring area, the respiratory frequency and sweating degree of the operator are extracted, and the respiratory frequency and sweating degree are combined to generate a bivariate observation sequence; the Euclidean distance is used as the similarity measure to iteratively optimize the cluster center, the bivariate observation sequence is subjected to cluster analysis, and the physiological behavior mode category of the bivariate observation sequence is divided according to the cluster analysis result; based on the physiological behavior mode category, a behavior coupling model is established using the Bayesian prior distribution technology, and the Markov chain Monte Carlo technology is used to sample the posterior distribution of the behavior coupling model; the time dynamic parameters are estimated according to the sampling result, the respiratory frequency and sweating degree of the operator within a target time period are predicted, the respiratory frequency and sweating degree are converted into the total volume of respiratory gas and the amount of skin sweat secretion by integrating the change, the physiological output sequence changing with time is generated, and the molecular contamination load is obtained.

[0052] It should be explained that in the process of obtaining the molecular contamination load, the working behavior of the operator in the monitoring area is simulated, a fine-grained task behavior model is constructed, specific operation tasks are generated according to the working content of the actual equipment in the area, the operation tasks are subdivided into standardized behavior action nodes such as carrying chemicals, operating valves, and detecting instruments, action operation parameters such as action duration, tools used, contact area, task frequency, etc. are set for each node to form execution condition nodes, and at the same time, a behavior rule editor is used to start a behavior tree creation wizard to construct a blank behavior tree framework containing a root node and multiple layers of sub-node containers. The rule editor writes behavior logic for each node to obtain a runnable behavior model, which is bound to the entity operator simulation model to receive and execute operation instructions from the behavior rule engine, thereby reproducing the behavior path of the operator in the simulation environment; the path information after simulation running is used to acquire the behavior state of the operator under different spatial positions and operation conditions, and further extract key physiological parameters including respiratory frequency (such as 15 breaths per minute) and sweating degree (such as 150ml of sweat secretion per hour under high-intensity work), to generate a bivariate observation sequence.

[0053] To simulate the behavior-physiology coupling accurately, the Euclidean distance is used as the clustering basis, and K-means++ is used to optimize the initial center point. The clustering analysis is performed to divide the operators into different physiological behavior patterns, such as high metabolic activity type, moderate interaction type, and low exposure type. At the same time, the Bayesian prior distribution is established based on the clustering label, and the Markov Chain Monte Carlo method is used to iteratively sample the posterior distribution to estimate the time dynamic parameters. Combined with the integral conversion model, the time dynamic respiratory frequency (such as gradually increasing from 12 times per minute to 18 times per minute) and the sweating degree (such as significantly increasing at the 30th minute of work) are converted into the total respiratory gas volume (for example, 2.4 cubic meters) and the sweat secretion amount (for example, 1.2 liters), and the physiological output sequence covering the entire target time length is constructed. This sequence is the molecular contamination load of the operator.

[0054] At the same time, it is assumed that a dangerous chemical plant is 15m long, 10m wide, and has 6 running devices, each device corresponding to an operator. According to the operation plan, 6 tasks (such as loading, stirring, sampling, etc.) are generated, which are decomposed into 18 behavior action nodes, with an average duration of 8 minutes. The rule editor defines that the sampling node needs to manually open the valve and approach the device more than 0.5m, which belongs to the high contact and high metabolic task. After the behavior tree is created, 6 entity operator simulation models are bound, the behavior path simulation is run, and the simulation data extraction shows that operator A is in a high intensity state from the 10th to 30th minute, with a stable respiratory frequency of 18 times per minute and a sweating rate of about 160ml / h. The two-variable sequence of operator A is synthesized, and it is divided into high, medium, and low metabolic patterns through K=3 clustering analysis. Operator A is classified into the high metabolic group. At the same time, the behavior coupling model estimates that the total respiratory gas volume of operator A within 60 minutes is 2.6m 3 , and the total sweat secretion amount is 1.5L, and the pollution load is output.

[0055] Step S2, extract the molecular contaminant concentration data in the historical operation process in the monitoring area, and couple the corresponding number of operators. Based on the coupling result, the causal relationship between the molecular pollution load and the molecular contaminant concentration is obtained.

[0056] In one embodiment, the historical molecular pollutant concentration data in the monitoring area during the historical operation process is extracted, and the corresponding number of operators is coupled, and the causal relationship between the molecular pollution load and the molecular pollutant concentration is obtained based on the coupling result, including: collecting the historical pollutant concentration data and the number of operators in the monitoring area, and dividing the historical molecular pollutant concentration data and the number of operators into several time periods, judging the fluctuation characteristics of the historical molecular pollutant concentration data and the change mode of the operators in each time period, and generating a time series data set; based on the time series data set and the molecular pollution load, a neural network is used for time series modeling processing to capture the long-term dependence relationship between the historical molecular pollutant concentration change and the molecular pollution load, and generate a causal network; the causal network is subjected to sensitivity analysis to evaluate the contribution degree of the number of operators in the causal network, and the causal network is optimized according to the contribution degree to capture the potential causal relationship between the time series data set and the molecular pollution load.

[0057] It needs to be explained that in the process of capturing the potential causal relationship between the time series data set and the molecular pollution load, the historical molecular pollutant concentration data and the number of operators in the monitoring area are collected, specifically including the pollutant concentration change at each time point in the area (such as the concentration of sulfur dioxide and nitrogen oxides per hour) and the number of operators in each period (such as the number of workers per hour), and the historical data is divided by time period, for example, divided by hour or by shift, to obtain the fluctuation characteristics of the pollutant concentration and the change mode of the number of operators in each time period. Through analysis of the pollutant concentration and the number of operators in the time period, a time series data set is generated, which includes the number of operators and the corresponding pollutant concentration data in each time period, reflecting the change trend of the pollutant concentration under different operation intensities and the influence of the operators on the pollutant concentration.

[0058] On the basis of the trend of pollutant concentration and the influence of operating personnel on pollutant concentration, a neural network is used for time series modeling to capture the long-term dependence between pollutant concentration changes and the number of operating personnel. The model learns the time series characteristics of historical data and captures the dynamic change pattern between pollutant concentration and the number of operating personnel. During the model training process, the input is the number of operating personnel and the pollutant concentration in the historical time period, and the output is the predicted future pollutant concentration or molecular pollution load value. Through the modeling results, potential causal relationships in the historical data can be identified to generate a causal network. The causal network can intuitively show the interaction between the number of operating personnel and the pollutant concentration, and reveal the mechanism of operating personnel in the process of pollutant concentration change. After generating the causal network, sensitivity analysis is performed to evaluate the impact of different variables (such as the number of operating personnel) on each causal relationship in the causal network, and determine the contribution of the number of operating personnel to the change in pollutant concentration. Specifically, by modifying the value of the number of operating personnel or other related variables, the change in pollutant concentration in the causal network is observed, and the optimized causal network can more accurately capture the potential causal relationship between time series data and molecular pollution load.

[0059] Suppose the sulfur dioxide (SO2) concentration and the number of operating personnel in a chemical plant workshop are monitored as follows. Through historical data collection, the hourly variation records of sulfur dioxide concentration (unit: ppb) and the hourly variation records of the number of operating personnel (unit: people) in the workshop for a certain period of time (such as 3 months) are as follows:

[0060] From 8:00 to 9:00 on January 1st, the sulfur dioxide concentration is 120 ppb, and the number of operating personnel is 5;

[0061] From 9:00 to 10:00 on January 1st, the sulfur dioxide concentration is 130 ppb, and the number of operating personnel is 8;

[0062] From 10:00 to 11:00 on January 1st, the sulfur dioxide concentration is 125 ppb, and the number of operating personnel is 6;

[0063] After the data is divided into time periods, a time series dataset is obtained, where each time period includes the corresponding sulfur dioxide concentration and the number of operating personnel. For example, in a certain time period, the change in the number of operating personnel may cause fluctuations in the concentration of pollutants. By analyzing the time series data, a dataset is generated for neural network learning. The neural network model (such as LSTM) learns the time series features to capture the long-term dependence relationship between the number of operating personnel and the concentration of pollutants. The model predicts that when the number of operating personnel increases from 5 to 8, the sulfur dioxide concentration increases by an average of 10 ppb. Based on the training results, a causal network is generated, where the relationship between the number of operating personnel and the concentration of sulfur dioxide is quantified, and the impact of the number of operating personnel on the concentration of pollutants is evaluated. By modifying the value of the number of operating personnel (for example, increasing the number of operating personnel from 5 to 10), it is found that the change in the concentration of sulfur dioxide is significantly increased, which indicates that the number of operating personnel plays an important role in the fluctuation of the concentration of sulfur dioxide. Based on the analysis results, the causal network is optimized to remove unimportant variables (such as temperature), and finally an accurate causal network is generated, making the prediction of the concentration of sulfur dioxide more accurate.

[0064] Further, through the time series modeling and sensitivity analysis of the neural network model, the causal impact of the number of operating personnel on the concentration of pollutants can be analyzed. The causal network can help decision-makers more accurately predict the changes in the concentration of pollutants under different operating intensities and personnel distribution, and thus provide strong data support for optimizing the layout of monitoring sampling points.

[0065] Step S3, obtain the number and distribution of operating personnel in the monitoring area in the target monitoring period, and combine the number of operating personnel with the causal relationship to predict the concentration of molecular pollutants in the monitoring area and determine the multi-pollutant source singular point.

[0066] It needs to be explained that the number of operators in the monitoring area in the target monitoring period and the distribution position are obtained, for example, the operator distribution information of a certain monitoring area from 8:00 to 10:00 in the morning can be obtained through the scheduling system of the factory, and the personnel distribution of each post and work area, assuming that there are 10 operators in the monitoring area in a certain period, distributed in different work areas, of which 3 are in the area close to the pollution source, 5 are in the middle area, and 2 are in the area away from the pollution source. Through the distribution information, the mutual relationship between the operators and the pollution source can be analyzed, and further input data for the prediction model of the pollutant concentration is provided. The number of operators and the causal relationship are combined to predict the concentration of the molecule pollutant in the monitoring area. In the prediction process, the previously established causal relationship model is used, which has captured the causal relationship between the number of operators, the distribution and the concentration of pollutants by analyzing the relationship between historical data and operation behavior and the concentration of pollutants. Specifically, the causal relationship model predicts the concentration of pollutants in the area according to the distribution of operators in different areas (such as whether the operators are concentrated near the pollution source), for example, if more operators are concentrated in the high pollution source area, the concentration of pollutants in that area may be relatively high, while the concentration of pollutants in other areas is relatively low. Based on this causal relationship model, the concentration of pollutants in different areas in the target period is calculated in real time, so as to determine the distribution of the concentration of pollutants in the monitoring area.

[0067] After determining the concentration of pollutants, through analysis of the concentration of pollutants in different areas, multiple pollution source singular points are identified. The multiple pollution source singular points refer to the abnormal fluctuation or concentration of pollutants under certain special environmental conditions. The area may become a key point for pollution control, for example, some areas may have equipment failure, too high operator density, or poor local ventilation, etc., resulting in excessive concentration of pollutants, becoming potential pollution source singular points. Through comprehensive analysis of the concentration of pollutants and the distribution of operators, these singular points are identified, providing a reference for subsequent sampling point optimization.

[0068] Suppose in a production workshop of a chemical plant, the target monitoring period is one day of working time (8:00-18:00), and there are 12 operators in the monitoring area during this period. Through the scheduling task, it is known that the operators are distributed in different areas, 3 in the high-pollution source area, 5 in the medium-pollution source area, and 4 in the low-pollution source area. The causal model shows that for each additional operator near the pollution source, the pollution concentration will increase by 15 ppb (unit: pollution concentration). According to this causal relationship, first calculate the pollution concentration of each area. Assuming that during the period from 8:00 to 10:00, there are 3 operators in the high-pollution source area, the pollution concentration in this area is 45 ppb (3 people x 15 ppb). In the medium-pollution source area, the total pollution concentration of 5 operators is 75 ppb (5 people x 15 ppb). In the low-pollution source area, the pollution concentration of 4 operators is 60 ppb (4 people x 15 ppb). After calculating the pollution concentration distribution of each area, the pollution concentration in the medium-pollution source area reaches 75 ppb, exceeding the standard limit value (such as 50 ppb), so it is identified as a pollution source anomaly point.

[0069] Through real-time prediction and analysis, not only can potential high-pollution risk areas be discovered in advance, but also decision-makers can be provided with the basis for optimizing monitoring sampling points. For example, the area with the highest pollution concentration in the monitoring area is the heavy pollution source area, so the focus of the monitoring sampling point can be placed in this area, and the sampling frequency can be increased to monitor and control the pollution concentration in this area in a timely manner to ensure that it does not exceed the standard.

[0070] Step S4, generate an initial distribution map of monitoring sampling points based on the distribution of multiple pollution source anomaly points and operators, and optimize the initial distribution map to determine the final monitoring sampling points according to the application cost under the condition of the initial distribution map.

[0071] In one embodiment, generating an initial distribution map of monitoring sampling points based on the distribution of multiple pollution source anomaly points and operators, and optimizing the initial distribution map to determine the final monitoring sampling points according to the application cost under the condition of the initial distribution map includes: determining the upper limit value of the molecular pollution concentration of the monitoring area based on the multiple pollution source anomaly points, and determining the maximum number value of the monitoring sampling points according to the upper limit value and the activity intensity of the distribution position of the operators; constructing a bounding box using axis alignment and directional bounding box technology according to the maximum number value of the monitoring sampling points and the layout state of the monitoring area, and generating an initial monitoring sampling point distribution map based on the bounding box; obtaining the purchase price of the monitoring sampling instrument, and calculating the application cost generated by the initial monitoring sampling point distribution map according to the purchase price, establishing a minimum objective function based on the application cost, and optimizing the initial distribution map to determine the final monitoring sampling points.

[0072] In one embodiment, according to the maximum number of monitoring sampling points and the layout state of the monitoring area, an axis-aligned and oriented bounding box technique is used to construct a bounding box, and an initial monitoring sampling point distribution map is generated based on the bounding box, which includes: obtaining the maximum number of monitoring sampling points and the layout state of the monitoring area, and determining the optimal coordinate axis direction by using the axis-aligned bounding box and the oriented bounding box construction technique in combination with the layout state of the monitoring area; constructing a bounding box that completely covers the monitoring area based on the optimal coordinate axis direction, and projecting the monitoring sampling points to the nearest plane of the bounding box according to the layout state of the monitoring area to output the monitoring sampling point positions; determining the distance between the monitoring sampling points and the entities in the monitoring area according to the monitoring sampling point positions, and eliminating the monitoring sampling points with a distance higher than a distance threshold, while re-projecting the eliminated monitoring sampling points; until the difference between the monitoring sampling point positions and the entities in the monitoring area is less than or equal to the distance position, an initial monitoring sampling point distribution map is obtained.

[0073] In one embodiment, an initial distribution map of monitoring sampling points is generated based on the singular points of multiple pollution sources and the distribution positions of operators, and the final monitoring sampling points are determined by optimizing the initial distribution map according to the application cost, so as to ensure that the key pollution sources and operating behaviors in the monitoring area are fully covered while reducing unnecessary monitoring points, improving the efficiency and cost-effectiveness of monitoring. By determining the upper limit value of the pollution concentration in the monitoring area based on the singular points of multiple pollution sources, and combining the distribution density and activity intensity of operators to determine the maximum number of monitoring sampling points, the upper limit value of the pollution concentration (such as the maximum value of the pollution concentration near the pollution source, for example, 80 ppb) can be determined through causal relationship analysis and identification of pollution source singular points. According to this concentration upper limit value and the distribution of operators in different areas (for example, the personnel concentration in a certain area is higher, resulting in higher pollution concentration in that area), the number of required monitoring sampling points can be calculated.

[0074] According to the maximum number of monitoring sampling points and the layout state of the monitoring area, an axis-aligned and oriented bounding box technique (AABB and OBB technique) is used to construct a bounding box to generate an initial monitoring sampling point distribution map for the monitoring area in a reasonable manner. Axis-aligned bounding box and oriented bounding box can efficiently perform spatial layout. When the geometric form of the monitoring area is complex, the oriented bounding box can more accurately enclose irregular areas, thereby reducing redundant sampling points and ensuring sufficient coverage of key areas. Assuming that a certain monitoring area presents an irregular polygon shape, the AABB method may result in unnecessary monitoring points, while the OBB technique can obtain a more reasonable monitoring point distribution, reducing redundant monitoring.

[0075] Suppose a chemical plant has a rectangular workshop with a length of 100 meters and a width of 50 meters, and the target monitoring period is 24 hours. In the workshop, there are 3 main pollution sources, and during the monitoring period, there are 30 people distributed in different areas. Suppose pollution source 1 and pollution source 2 are relatively close, and pollution source 3 is located at the other end of the workshop. According to historical data, the upper limit of the pollutant concentration in the chemical reactor area is 80 ppb, the concentration in the evaporator area is 75 ppb, and the concentration in the exhaust gas pipeline area is 50 ppb.

[0076] According to the distribution density of the operators, suppose there are 12 operators in the chemical reactor area, 8 in the evaporator area, and 10 in the exhaust gas pipeline area. According to this distribution, it is calculated that more sampling points are needed to cover the high pollution source areas such as the chemical reactor and the evaporator. Assuming the maximum number of monitoring sampling points is 30, the sampling points are distributed in different pollution source areas, and the axis alignment and directional bounding box technology is used for layout optimization to construct a bounding box that can rationalize the irregular layout in the workshop, optimize the sampling points to the high pollution area, and effectively reduce the redundant points.

[0077] In one embodiment, the purchase price of the monitoring sampling instrument is obtained, and the application cost of the initial monitoring sampling point distribution diagram is calculated based on the purchase price. Based on the application cost, a minimum objective function is established to optimize the initial distribution diagram to determine the final monitoring sampling points, including: obtaining the purchase price of the monitoring sampling instrument, and calculating the purchase cost of the monitoring sampling instrument under the condition of the maximum number of monitoring sampling points, while predicting the deployment, operation and maintenance and power additional cost of the initial monitoring sampling point distribution diagram during the monitoring process; based on the additional cost and the purchase cost, a total application cost function is constructed, and the minimum total application cost is taken as the target, combined with the monitoring coverage rate and the deployment efficiency to construct a multi-objective function for optimizing the monitoring sampling point distribution; according to the multi-objective function, the monitoring sampling point distribution scheme is coded, and the monitoring sampling point distribution scheme is defined as a node, and the non-dominated sorting reverse learning differential evolution algorithm is used to determine the final monitoring sampling point.

[0078] In one embodiment, the monitoring sampling point distribution scheme is coded according to a multi-objective function, and the monitoring sampling point distribution scheme is defined as a node, and the final monitoring sampling point is determined by using a non-dominated sorting reverse learning differential evolution algorithm, including: coding the monitoring sampling point distribution scheme into a node representation based on the multi-objective function, monitoring the configuration of the sampling point by using the node representation, and performing differential mutation processing on the node by using the differential evolution algorithm to generate a mutation vector; performing a crossover operation on the mutation vector according to a crossover probability to generate a test individual, and calculating the fitness value of each test individual, and calculating the performance of the corresponding node under the multi-objective function based on the fitness value; selecting the optimal solution from the test individual based on the performance calculation result and the greedy selection strategy; and outputting a Pareto optimal solution set according to the optimal solution, selecting a monitoring sampling point distribution scheme with the lowest total cost, the highest monitoring coverage rate and the highest deployment efficiency from the Pareto optimal solution set as the final monitoring sampling point.

[0079] In one embodiment, selecting the optimal solution from the test individual based on the performance calculation result and the greedy selection strategy includes: sorting the test individuals according to their performance on the multi-objective function, comparing the comprehensive performance of the test individuals on the multi-objective function to determine the existence of a dominance relationship; performing hierarchical sorting processing on the test individuals according to the existence of the dominance relationship, dividing the Pareto level based on the hierarchical sorting result, and gradually selecting the test individual with the best performance from the Pareto level by using the greedy selection strategy to output the optimal solution.

[0080] It needs to be explained that the purchase price of the monitoring sampling instrument is obtained, and it is assumed that the purchase price of each sampling instrument is 2000 yuan. Based on the area of the monitoring area and the number of required sampling points, the purchase cost of the sampling instrument is calculated, and it is assumed that the maximum number of sampling points in the monitoring area is 30. According to the purchase price of each sampling instrument, the total purchase cost of the sampling instrument is preliminarily calculated as: purchase cost = 30 * 2000 = 60000 yuan.

[0081] In addition to the purchase cost, the deployment, operation and maintenance and power additional cost of the monitoring sampling point in the monitoring process also need to be predicted. The cost includes the daily maintenance cost of the sampling instrument, the power consumption cost (such as 10 yuan of power cost per instrument per day), and the operation and maintenance cost of point deployment, debugging and data analysis. It is assumed that the annual operation and maintenance cost of each sampling point in the operation stage is 1500 yuan, and the power additional cost is 10 yuan of power cost per sampling instrument per month. Then the annual operation and maintenance and power additional cost of the entire monitoring system is: additional cost = 30 * (1500 + 10 * 12) = 30 * 1740 = 52200 yuan.

[0082] Based on the purchase cost and the additional cost, a total application cost function is constructed, which can be expressed as: total application cost = purchase cost + additional cost = 60000 + 52200 = 112200 yuan / year.

[0083] Based on the total application cost required for obtaining the monitoring sampling point distribution, and the need to optimize these costs, in this optimization process, the goal is not only to minimize the total application cost, but also to consider the monitoring coverage and deployment efficiency, a multi-objective function is constructed to consider these goals at the same time, which combines the monitoring coverage (the range of pollution sources covered by each sampling point) and the deployment efficiency (the density and cost of monitoring point layout) to optimize the sampling point distribution, the monitoring sampling point distribution scheme is encoded as a node, and the differential evolution algorithm is applied for optimization, the differential evolution algorithm is a global optimization algorithm based on population, which is suitable for handling multi-objective optimization problems, in the algorithm, each individual represents a monitoring sampling point distribution scheme, and all individuals form a population, the fitness value of each individual is related to its performance in the multi-objective function.

[0084] Specifically, the differential evolution algorithm first performs mutation processing on each node to generate a mutation vector, the generation of the mutation vector is achieved by selecting three random individuals from the current population, calculating the difference vector of the individuals and adding it to the current individual, assuming that the current individual represents the layout of a sampling point, three individuals are selected to calculate their differences and generate a mutation vector, and then a new trial individual is formed, according to the crossover probability, the mutation vector is crossed to generate a new trial individual, which represents a new monitoring sampling point distribution scheme.

[0085] The fitness value of each trial individual is calculated, and its performance in the multi-objective function is evaluated, the calculation of the fitness value will consider the total application cost, monitoring coverage and deployment efficiency, if a scheme has a lower total application cost, but its ability to cover high pollution sources is weak, its fitness value may be lower, on the contrary, if a scheme can effectively cover all pollution sources and has high deployment efficiency, its fitness value is higher, based on the fitness value, the optimal solution is selected from the trial individuals through the greedy selection strategy, the greedy selection strategy first sorts the trial individuals, compares their advantages and disadvantages in multiple objectives, and selects the best individual from the trial individuals according to the priority, the advantages and disadvantages of the trial individuals are judged by their comprehensive performance in multiple objectives, the performance of each individual in all objective functions is compared, the individuals with strong dominance relationship are selected, and these individuals are put into the next round of optimization, the Pareto optimal solution set is output, which means that in multi-objective optimization, the solution of other objectives cannot be improved by adjusting the value of any one objective, finally, the monitoring sampling point distribution scheme with the lowest total cost, the highest monitoring coverage and the highest deployment efficiency is selected as the final monitoring scheme, in this way, the optimal sampling point layout scheme with comprehensive performance can be obtained in the framework of multi-objective optimization.

[0086] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the present embodiment can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software manner depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0087] Although the specific embodiments of the present application are described above in combination with the drawings, it is not a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications or variations made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the scope of protection of the present application.

Claims

1. An AMC sampling point distribution optimization method based on multi-pollution source singularity point fusion, characterized in that, include: Based on the equipment's work content in the monitoring area, the operator's work tasks are generated, the work tasks are decomposed into behavioral action nodes, and the action operation parameters are set for the behavioral action nodes to obtain the execution condition nodes. The behavior tree creation wizard is launched based on the autonomous behavior rule editor to generate a blank behavior tree framework containing several root nodes and child node containers. The rule generator is used to define the behavior logic for the root nodes and child nodes. A blank behavior tree is obtained based on the behavioral logic definition results. Execution condition nodes are added to the blank behavior tree nodes to extract operation instruction information. A behavior rule engine is generated based on the extraction results. The behavior rule engine is associated with the simulation model of the actual operator, and operation instruction information is sent to the simulation model of the actual operator to determine the behavior path of the actual operator in the virtual monitoring area. In the simulation environment, a blank behavior tree structure is run according to the behavior path of the actual operator, and the working behavior of the operator in the virtual monitoring area is obtained based on the running results. The respiratory rate and sweating level of operators under working conditions are obtained, a behavioral-physiological coupling model is constructed, and the respiratory volume and skin secretion volume generated by operators within the target duration are predicted to obtain the molecular pollution load. Extract the molecular pollutant concentration data from the historical operation process within the monitoring area and couple it with the corresponding number of operators. Based on the coupling results, obtain the causal relationship between molecular pollutant load and molecular pollutant concentration. The number and distribution of operators in the monitoring area during the target monitoring period are obtained, and the number of operators is combined with causal relationships to predict the concentration of molecular pollutants in the monitoring area and identify singularities of multiple pollution sources. An initial distribution map of monitoring sampling points is generated based on the singularities of multiple pollution sources and the distribution locations of operators. The final monitoring sampling points are determined by optimizing the initial distribution map based on the application cost under the conditions of the initial distribution map.

2. The AMC sampling point distribution optimization method based on multi-pollution source singularity fusion according to claim 1, characterized in that, The process involves acquiring the operator's respiratory rate and sweating level under working conditions, constructing a behavioral-physiological coupling model, predicting the operator's respiratory volume and skin secretion volume within a target duration, and obtaining the molecular pollution load, including: Based on the work behavior within the virtual monitoring area, the respiratory rate and sweating level of the operators are extracted, and the respiratory rate and sweating level are combined to generate a bivariate observation sequence; Cluster centers were iteratively optimized using Euclidean distance as a similarity metric. Cluster analysis was performed on bivariate observation sequences, and the physiological behavior pattern categories of the bivariate observation sequences were classified based on the cluster analysis results. A behavior coupling model is established based on physiological behavior pattern categories using Bayesian prior distribution technology, and Markov chain Monte Carlo technology is used to sample the posterior distribution of the behavior coupling model. Based on the sampling results, time dynamic parameters are estimated, and the respiratory rate and sweating degree of the operator within the target time period are predicted. The respiratory rate and sweating degree are converted into the total volume of respiratory gas and the amount of skin sweat secretion through integral change, generating a physiological output sequence that changes over time, and thus obtaining the molecular pollution load.

3. The AMC sampling point distribution optimization method based on multi-pollution source singularity fusion according to claim 1, characterized in that, The process of extracting historical molecular pollutant concentration data within the monitoring area and coupling it with the corresponding number of operators, and obtaining the causal relationship between molecular pollutant load and molecular pollutant concentration based on the coupling result, includes: Collect historical pollutant concentration data and the number of operators within the monitoring area, divide the historical molecular pollutant concentration data and the number of operators into several time periods, determine the fluctuation characteristics of historical molecular pollutant concentration data and the change pattern of operators within each time period, and generate a time series dataset. Based on time series datasets and molecular pollution loads, neural networks are used for time series modeling to capture the long-term time-series dependence between historical molecular pollutant concentration changes and molecular pollution loads, generating causal networks. Sensitivity analysis was performed on the causal network to assess the contribution of the number of operators to the causal network. The causal network was then optimized based on the contribution to capture the potential causal relationship between the time series dataset and the molecular pollution load.

4. The AMC sampling point distribution optimization method based on multi-pollution source singularity fusion according to claim 1, characterized in that, include: The upper limit of molecular pollutant concentration in the monitoring area is determined based on singularities from multiple pollution sources. The maximum number of monitoring sampling points is determined based on the upper limit and the activity density of the operators' distribution locations. Based on the maximum number of monitoring sampling points and the layout of the monitoring area, a bounding box is constructed using axis alignment and oriented bounding box techniques, and an initial monitoring sampling point distribution map is generated based on the bounding box. Obtain the purchase price of the monitoring sampler, calculate the application cost generated by the initial monitoring sampling point distribution map based on the purchase price, establish a minimum objective function based on the application cost, optimize the initial distribution map, and determine the final monitoring sampling points.

5. The AMC sampling point distribution optimization method based on multi-pollution source singularity fusion according to claim 4, characterized in that, The process of constructing a bounding box using axis alignment and orientation bounding box techniques based on the maximum number of monitoring sampling points and the layout of the monitoring area, and generating an initial monitoring sampling point distribution map based on the bounding box, includes: The maximum number of monitoring sampling points and their layout status are obtained, and the optimal coordinate axis direction is determined by using axis-aligned bounding boxes and oriented bounding boxes to construct bounding boxes, combined with the layout status of the monitoring area. Construct a bounding box that fully covers the monitoring area based on the optimal coordinate axis direction, and project the monitoring sampling points onto the nearest plane of the bounding box according to the layout of the monitoring area, and output the position of the monitoring sampling points; The distance between the monitoring sampling point and the entity within the monitoring area is determined based on the location of the monitoring sampling point, and monitoring sampling points whose distance is higher than the distance threshold are removed. At the same time, the removed monitoring sampling points are reprojected. The initial monitoring sampling point distribution map is obtained when the difference between the location of the monitoring sampling point and the distance to the entity within the monitoring area is less than or equal to the distance to the location.

6. The AMC sampling point distribution optimization method based on multi-pollution source singularity fusion according to claim 4, characterized in that, The term includes: Obtain the purchase price of the monitoring sampler and calculate the acquisition cost of the monitoring sampler under the condition of the maximum number of monitoring sampling points. At the same time, predict the deployment, operation and maintenance and power surcharge of the initial monitoring sampling point distribution map during the monitoring process. A total application cost function is constructed based on additional costs and acquisition costs. With the goal of minimizing the total application cost, a multi-objective function is constructed to optimize the distribution of monitoring sampling points by combining monitoring coverage and deployment efficiency. The monitoring sampling point distribution scheme is encoded based on a multi-objective function, and the monitoring sampling point distribution scheme is defined as a node. The final monitoring sampling points are determined by using a non-dominated sorting backward learning differential evolution algorithm.

7. The AMC sampling point distribution optimization method based on multi-pollution source singularity fusion according to claim 6, characterized in that, The term includes: The monitoring sampling point distribution scheme is encoded into a node representation based on a multi-objective function. The node representation is used to configure the monitoring sampling points, and the differential evolution algorithm is used to perform differential mutation processing on the nodes to generate mutation vectors. Crossing the mutation vectors according to the crossover probability generates experimental individuals, and the fitness value of each experimental individual is calculated. Based on the fitness value, the performance of the corresponding node under the multi-objective function is calculated. The optimal solution is selected from the experimental individuals based on the performance calculation results and a greedy selection strategy; Based on the optimal solution, output the Pareto optimal solution set, and select the monitoring sampling point distribution scheme with the lowest total cost, highest monitoring coverage and highest deployment efficiency from the Pareto optimal solution set as the final monitoring sampling points.

8. The AMC sampling point distribution optimization method based on multi-pollution source singularity fusion according to claim 7, characterized in that, The selection of the optimal solution from experimental individuals based on performance calculation results and a greedy selection strategy includes: Based on the performance of nodes under multi-objective functions, the experimental individuals are ranked according to their performance on multiple objectives, and the overall performance of the experimental individuals under multiple objectives is compared to confirm the existence of the dominance relationship. The experimental individuals are stratified and ranked according to the existence of dominance relationships. Pareto levels are defined based on the stratification and ranking results. A greedy selection strategy is then used to select the best-performing experimental individual from the Pareto levels step by step, and the optimal solution is output.

Images