A method and system for controlling perchlorate point source pollution driven by environmental variables

Abstract

The invention discloses a method and system for controlling perchlorate point source pollution driven by environmental variables, and relates to the technical field of water source pollution control; the method comprises the following steps: collecting environmental variable data; identifying high-risk pollution areas: constructing a water source pollution risk prediction model based on key environmental variables, predicting the probability of water source pollutants exceeding the standard, and identifying high-risk pollution areas; calculating the pollution state of drinking water sources: constructing a physical diffusion process simulation model, simulating the diffusion path of pollutants in rivers in different seasons, using the physical diffusion process simulation model to simulate the diffusion path of point source pollutants in high-risk pollution areas, and calculating the pollutant concentration of drinking water sources; determining the optimal pollution reduction strategy: constructing a pollution reduction optimization model driven by reinforcement learning, and encouraging the pollution reduction optimization model to find the optimal pollution reduction strategy; formulating a comprehensive pollution reduction strategy. The invention comprehensively considers various factors, formulates a comprehensive pollution reduction strategy, and improves the emission reduction effect.

Description

Technical Field

[0001] The invention belongs to the technical field of water source pollution control, and in particular relates to a method and system for controlling perchlorate point source pollution driven by environmental variables. Background Art

[0002] Perchlorate is one of the main inorganic pollutants in the environment and is widely present in surface water, groundwater and soil. Perchlorate in the environment mainly comes from the military, agriculture and industry. In industrial production, perchlorate is mainly used as an additive in leather processing, rubber manufacturing, paint production, etc. The huge demand in industrial production has caused artificial perchlorate produced by factories. Accompanied by air flow, water migration and biological enrichment, the environmental pollution problem of perchlorate is serious. Especially after perchlorate enters the water body, its diffusion is affected by the internal flow of the water source. How to monitor water source pollution is the difficulty of current environmental governance. Existing water source pollution monitoring technologies mainly include the following aspects:

[0003] Monitoring and sampling: By setting up monitoring points or monitoring stations at water sources, water samples are collected regularly, or automated sensors are used for real-time monitoring, and the water samples are sent to laboratories for analysis of the types and concentrations of pollutants. Monitoring and sampling can obtain information on pollution concentrations with high accuracy. However, the frequency of monitoring and sampling is limited, and it is difficult to provide real-time and continuous data. In addition, monitoring and sampling methods can only obtain information at the sampling point and cannot reflect the situation of the entire water body. This is because the distribution of pollutants in the water body is uneven and will change with changes in factors such as water flow and wind force.

[0004] Remote sensing monitoring: Obtain spectral information of water bodies through remote sensing equipment such as satellites or drones, analyze spectral characteristics of different bands to identify and quantify some pollutants in the water. However, the accuracy of remote sensing monitoring technology is limited, and it cannot detect all types of pollutants in water bodies. It can only identify some pollutants with obvious characteristics. It is also greatly affected by weather and equipment resolution, and it is difficult to penetrate deep into the water body or detect pollutants without obvious spectral characteristics.

[0005] Among the existing water pollution monitoring technologies, monitoring and sampling mainly rely on regular sampling and laboratory analysis, which makes it difficult to provide real-time and continuous data. Although remote sensing technology can cover a large range of water bodies, its accuracy is limited and it cannot detect all types of pollutants. These water pollution monitoring technologies have greatly delayed the subsequent water pollution control. In addition, the existing pollution reduction measures lack systematic and dynamic adjustment capabilities in terms of industrial upgrading, process optimization, and water treatment in water plants.

[0006] In the process of water pollution control, the main method for optimizing water pollution control is emission reduction based on emission standards / water quality standards. The formulation of emission standards / water quality standards is based on methods such as the dilution multiple method, the pollutant reduction coefficient method, and the water body pollution assimilating capacity method to evaluate the water body's pollution assimilating capacity and determine the maximum concentration of pollutants that the water body can accept, so as to formulate reasonable water quality standards. The formulation of existing emission standards and water quality standards is relatively static, and the standard formulation process mainly relies on the dilution multiple method, the pollutant reduction coefficient method, and the water body pollution assimilating capacity method. These methods are based on fixed assessment models, fail to adjust in combination with real-time monitoring data and dynamic models, and fail to fully consider the spatiotemporal dynamic changes and cumulative effects of pollutants, resulting in a lack of flexibility and timeliness in practical applications.

[0007] In summary, existing technologies lack dynamic adaptability and comprehensive optimization capabilities when dealing with complex pollution sources and rapidly changing environmental conditions, and are unable to effectively achieve precise and efficient pollution reduction.

[0008] The present invention provides a method and system for controlling perchlorate point source pollution driven by environmental variables, which improves the timeliness and continuity of data through real-time monitoring and dynamic analysis of water source pollution, and combines multi-source data fusion technology to comprehensively utilize remote sensing data and ground monitoring data to improve the accuracy and comprehensiveness of pollutant detection. It also includes a system that integrates environmental variables, industrial emission characteristics, and hydrological diffusion, comprehensively considers various factors, formulates a comprehensive pollution reduction strategy, and improves the emission reduction effect. Summary of the invention

[0009] The purpose of the present invention is to provide an environmental variable driven perchlorate point source pollution control method and system to solve the problems of insufficient real-time monitoring and sampling in the prior art proposed in the above background technology, limited accuracy of remote sensing monitoring, insufficient optimization of pollution reduction measures and lack of comprehensive emission reduction strategies.

[0010] To achieve the above object, the present invention adopts the following technical solutions:

[0011] The first aspect of the present invention provides a method for controlling perchlorate point source pollution driven by environmental variables, comprising the following steps:

[0012] S1, collect environmental variable data;

[0013] Collect environmental variable data related to water source pollution in the target area; environmental variable data include pollution information data, climate data, geographical environment data, hydrological data and industrial enterprise characteristic data;

[0014] S2. Identify high-risk pollution areas;

[0015] Develop a water source pollution risk prediction model based on machine learning using environmental variable data, determine the key environmental variables for predicting water source pollutants exceeding the standard, and build a water source pollution risk prediction model based on the key environmental variables to predict the probability of water source pollutants exceeding the standard; use the water source pollution risk prediction model to identify the target area and obtain the high-risk pollution area within the target area;

[0016] S3. Calculate the pollution status of drinking water sources;

[0017] A physical diffusion process simulation model is constructed based on hydrological variables to simulate the diffusion paths of pollutants in rivers in different seasons. The physical diffusion process simulation model is used to simulate the diffusion paths of point source pollutants in high-risk pollution areas, calculate the pollutant concentrations of drinking water sources in target areas, and dynamically simulate the pollution status of drinking water sources.

[0018] S4. Determine the optimal pollution reduction strategy;

[0019] The key environmental variables in S2 and the hydrological variables in S3 are used as the environmental states of the pollution reduction optimization model, and the reward function is designed to build a pollution reduction optimization model driven by reinforcement learning; different pollution reduction optimization strategies are designed to motivate the pollution reduction optimization model to find the optimal pollution reduction strategy;

[0020] S5. Develop comprehensive pollution reduction strategies;

[0021] Combined with the output of the pollution reduction optimization model, a comprehensive pollution reduction strategy is formulated; and based on the real-time monitoring data and the calculation results of the physical diffusion process simulation model, the pollution reduction strategy is adjusted to achieve dynamic pollution management.

[0022] Preferably, the high-risk pollution area is identified in S2 as follows:

[0023] S201, pre-processing the environmental variable data, and classifying the pollution information data into two categories, namely, exceeding the standard and not exceeding the standard;

[0024] S202, based on the preprocessed environmental variable data, respectively training the support vector machine, random forest, Catboost and multi-layer perceptron algorithms;

[0025] S203, using precision, recall and F1 score as evaluation indicators, selecting the best performing model, and performing hyperparameter optimization on the selected model parameters through Bayesian optimization strategy;

[0026] S204, using the model's built-in feature importance assessment method, the SHAP value-based method, and the statistical t-test method to identify key environmental variables that predict excessive water source pollutants;

[0027] S205. Reconstruct a water source pollution risk prediction model based on key environmental variables and evaluate the water source pollution risk prediction model;

[0028] S206. Predict the probability of water source pollutants exceeding the standard based on the water source pollution risk prediction model and identify high-risk pollution areas.

[0029] Preferably, the pollution state of the drinking water source is calculated in S3 as follows:

[0030] S301. Analyze high pollution risk areas using GIS software to determine the source of drinking water sources in the target area;

[0031] S302, based on hydrological variables, the SCHISM model is used to simulate water flow conditions in the dry season and the rainy season, and a physical diffusion process simulation model is constructed;

[0032] S303, calibrating a physical diffusion process simulation model using actual observation data in the target area;

[0033] S304, the physical diffusion process simulation model simulates the physical diffusion of point source pollutants discharged by industrial enterprises in high-risk pollution areas, and quantifies the pollution status of drinking water sources. This process has two main functions: First, by simulating the diffusion path of point source pollutants through the physical diffusion process simulation model, the prediction results of water source pollution risk prediction models can be verified and calibrated to improve the accuracy and reliability of the model. Second, the diffusion path of point source pollutants in high-risk areas is quantified to provide corrections for the formulation of reinforcement learning pollution control strategies.

[0034] Preferably, the pollution reduction optimization model driven by reinforcement learning is constructed in S4, specifically as follows:

[0035] A reinforcement learning algorithm is used to take key environmental variables and hydrological variables as the environmental states of the pollution reduction optimization model. The reward function of the pollution reduction optimization model is designed, and the knowledge of hydrological dynamics and pollutant behavior obtained from the traditional process model is embedded in the learning process. The strategy selection is adjusted according to the feedback of the reward function to find the optimal pollution reduction strategy.

[0036] Preferably, the reward function is as follows:

[0037] The reward function uses the water source pollution risk prediction model as the basis for calculation, calculates the probability of exceeding the standard for different drinking water sources, and defines the number of affected people based on the probability of exceeding the standard and the population distribution map. Ultimately, the number of affected people caused by exceeding the drinking water source pollution is used as the reward benchmark.

[0038] Preferably, different emission reduction optimization strategies are designed in S4, specifically as follows:

[0039] The emission reduction optimization strategy includes industrial transformation of industries and renewal of water treatment equipment in high-risk pollution areas, and the implementation cost of the corresponding emission reduction optimization strategy is calculated and the strategy is adjusted accordingly; the industrial transformation of industries is determined according to the pollution degree of point source pollutants to reduce the emission of point source pollutants and reduce pollutants in drinking water sources; water treatment equipment is renewed at drinking water sources or high-risk pollution areas to purify and reduce pollutants in drinking water sources.

[0040] Preferably, the calculation results of the physical diffusion process simulation model in S5 are as follows:

[0041] The diffusion paths of point source pollutants in high-risk pollution areas under different emission reduction optimization strategies are simulated according to the physical diffusion process simulation model, and the pollutant concentrations in drinking water sources are determined as a correction to the output results of the pollution reduction optimization model. The effects of different emission reduction optimization strategies are evaluated, and the output of the pollution reduction optimization model is corrected according to the simulation results, where the correction depends on the identification of high-risk pollution areas, which are predicted and determined by the water source pollution risk prediction model.

[0042] The second aspect of the present invention provides a perchlorate point source pollution control system driven by environmental variables obtained according to the above method, comprising:

[0043] Environmental variable data acquisition module, which integrates multiple sensors and data collection devices to collect climate, geographical environment, hydrology and industrial enterprise data, as well as drinking water source pollution information data;

[0044] Water source pollution risk prediction model, which uses geographic environment and climate data as feature inputs, and uses drinking water source pollution information as the target variable to build a machine learning model to predict the probability distribution of drinking water source pollution and identify high-risk areas;

[0045] Physical diffusion process simulation model, which establishes diffusion models for dry season and rainy season respectively, is used to calculate the pollutant concentration of drinking water sources to evaluate the impact of point source pollutants on drinking water sources;

[0046] The pollution reduction optimization model integrates key environmental variables and hydrological variables as environmental states for reinforcement learning, implements knowledge-guided reinforcement learning, and uses the number of people affected by excessive drinking water source pollution as the main reward to determine the optimal pollution reduction strategy.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] (1) The present invention has enhanced real-time and comprehensiveness. Traditional water quality monitoring methods mainly rely on regular sampling and laboratory analysis, which is not only time-consuming but also unable to capture the dynamic changes of the environment in real time. In contrast, the present invention realizes the real-time and comprehensive collection of environmental and industrial data by integrating multiple high-precision sensors and data sources. This improvement not only provides more detailed environmental variables, but also enables the water source pollution risk prediction model to perform accurate risk assessment based on trends and patterns through real-time data analysis, greatly improving the accuracy of prediction and the ability to respond to pollution events.

[0049] (2) The present invention has detailed simulation and evaluation. The present invention introduces the consideration of seasonal characteristics and significantly improves the traditional simulation of physical diffusion process. By establishing a detailed hydrological diffusion model for the dry season and the rainy season, the present technology can more scientifically simulate the diffusion behavior of pollutants under different environmental conditions. The application of this method not only enhances the accuracy of the model, but also provides more accurate data support for the assessment of the impact of pollutants on drinking water sources, thereby making the response measures more effective.

[0050] (3) The present invention optimizes pollution reduction strategies. Compared with traditional static pollution control strategies, the reinforcement learning-driven pollution reduction optimization model of the present invention introduces a dynamic adjustment mechanism. Using real-time environmental data and pollutant concentration feedback, the module can optimize emission reduction measures in real time through continuous learning and strategy iteration. The flexibility and adaptability of this strategy significantly improve the efficiency of environmental management and the actual effect of emission reduction measures.

[0051] (4) The present invention realizes comprehensive decision support. The present invention uses a highly integrated modular design to closely combine data collection, risk prediction, physical simulation and emission reduction strategy optimization to build an advanced decision support system. This systematic integration optimizes the information flow and decision-making process, and greatly improves the scientificity and effectiveness of water source management measures compared to existing fragmented processing methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A system block diagram of a perchlorate point source pollution control system driven by environmental variables in the present invention;

[0053] Figure 2 It is a flow chart of the environmental variable data acquisition module in the present invention;

[0054] Figure 3 It is a flow chart of the water source pollution risk prediction model in the present invention;

[0055] Figure 4 It is a flow chart of the physical diffusion process simulation model in the present invention;

[0056] Figure 5 It is a flow chart of the pollution reduction optimization model in the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] Embodiment 1:

[0059] See also Figure 1-5 , a method for controlling perchlorate point source pollution driven by environmental variables, comprising the following steps:

[0060] Step 1: Collection of environmental variables and pollution data.

[0061] Relevant pollution information data: Collect detailed records of historical pollution incidents, including environmental reports published on official websites, data released by environmental monitoring stations, and research reports from non-official organizations. Conduct time series analysis on the collected data to identify the frequency, severity, and duration of pollution incidents.

[0062] Climate data: Use the WorldClim global climate database to obtain historical and predicted climate data, including temperature, rainfall, humidity and wind speed, and combine it with real-time meteorological data provided by national meteorological monitoring stations to automatically download and update data sets through the API interface.

[0063] Geological and soil property data: The geological structure, rock type and its distribution characteristics of a specific area are extracted from the LiMW geological database, and key parameters such as soil type, texture, and organic matter content are obtained using the HWSD2.0 soil database.

[0064] Hydrological data: mainly obtained through regional hydrological monitoring stations and regional hydrological cycle data provided by the Earth2observe database.

[0065] Terrain feature data: ArcGIS software was used to analyze factors such as terrain slope, elevation, and terrain undulation using a grid calculator through a global elevation TIF map.

[0066] Industrial enterprise characteristic data: Through the geocoding API of Baidu and Google Maps, the distribution of enterprises in key industrial areas is collected, and the kernel density tool in ArcGIS software is used to calculate the distribution density of enterprises and the minimum distance between different locations and related enterprises is calculated through Python code. Finally, a sensor network with geographical location tags is installed in the selected typical pollution areas to collect real-time data and integrate it with historical data, and visit enterprises to investigate enterprise scale, wastewater discharge and process links for input and calibration of hydrological diffusion models.

[0067] Step 2: Development of water source pollution risk prediction model.

[0068] First, the collected environmental variables and pollution information data were preprocessed. For the input features, Python code was used to perform one-hot encoding on the text features, outliers were removed from all features, missing values ​​were filled, and the data was standardized. The pollution data was also classified into two categories: "exceeding the standard" and "not exceeding the standard".

[0069] Secondly, determine the model. Use Python's Scikit-learn library in the local computing environment to train and compare the support vector machine, random forest, Catboost, and multilayer perceptron algorithms, use precision, recall, and F1 score as evaluation indicators, and select the best performing algorithm to perform hyperparameter optimization on the selected model parameters through the Bayesian optimization strategy, and ensure the generalization ability of the model through 5-fold cross validation.

[0070] After determining the optimal parameters of the model, the feature importance assessment phase was entered. The model's built-in feature importance assessment method, the SHAP value-based method, and the statistical t-test method were used to identify key environmental variables that have a significant impact on predicting drinking water source pollution exceeding the standard. These key variables were then used to re-model and save as a callable model.

[0071] Finally, the optimized model was called through window sliding technology combined with real-time data to draw a pollution hotspot distribution map in the selected area, effectively identifying typical pollution hotspot areas.

[0072] Step 3: Establish a physical diffusion process simulation model.

[0073] S3-1. Use GIS software to divide the region in detail and conduct regional analysis based on hydrological characteristics such as river flow direction and topography, as well as the distribution of drinking water sources.

[0074] First, load the geographic information of the drinking water sources in the region and use the buffer tool to create a protection area for each water source. Next, import a high-resolution elevation TIF map and apply GIS tools to deeply analyze the river flow direction and terrain features; use the flow direction analysis tool to identify the flow direction of major rivers, and use the slope analysis tool to evaluate how the terrain affects the flow path. Finally, combine these analysis results with the distribution map of rivers in the region to determine the source of water for each drinking water source.

[0075] S3-2. The SCHISM model is selected to simulate the water flow conditions in the dry and rainy seasons and the migration and diffusion of wastewater discharged by nearby factories and enterprises in the watershed.

[0076] First, the necessary input data are collected and organized, including water quality data, flow and initial concentration of pollutants in the area. Then, the boundary conditions of the upstream inlet and downstream outlet are accurately set in the SCHISM model to reflect the actual hydrological characteristics of the river. In the model configuration, the geometry of the water body, flow rate and other relevant parameters are adjusted according to the actual topography and hydrological conditions. In order to adapt to the impact of climate change, especially the differences between the dry season and the rainy season, two sets of diffusion models were established respectively, so as to better understand how seasonal changes affect the behavior and diffusion paths of pollutants.

[0077] S3-3. Model calibration.

[0078] After the initial establishment of the model, the model was calibrated using actual observation data in the region, and simulation parameters such as hydrodynamic parameters and pollutant diffusion coefficients were adjusted to ensure that the model output was consistent with the pollutant concentration data measured in the field, thereby improving the accuracy and reliability of the model. After the calibration process was ensured to be correct, the model was used to simulate the diffusion trend of pollutants in real time. These real-time simulation results were directly input into the reinforcement learning system, and the output results of the reinforcement learning model were fine-tuned.

[0079] Step 4: Establish a knowledge-guided reinforcement learning model.

[0080] S4-1. Environment configuration of reinforcement learning model.

[0081] Define the environmental state of the reinforcement learning model (reinforcement learning driven pollution reduction optimization model), including key environmental variables identified from pre-trained machine learning models, such as climate, distribution of related industrial enterprises, soil physical and chemical properties, and variables in the hydrological diffusion model such as river flow and topography. These variables provide the model with a context for decision-making. Next, design a reward function to quantify the effect of governance, using the number of people affected by excessive drinking water pollution in a typical area as the reward benchmark, with the goal of minimizing the affected population, thereby intuitively reflecting the direct effectiveness of the emission reduction strategy.

[0082] S4-2. Strategy selection and implementation.

[0083] The selection of strategies involves two aspects, including industrial transformation and the renewal of water treatment equipment, and the implementation cost of the corresponding strategies is calculated as needed. Industrial transformation: Analyze the pollution contribution of factories in the region, and shut down or transform high-pollution factories according to the degree of pollution. For factories with relatively complete infrastructure facilities, they are transformed into more environmentally friendly production lines through process optimization. Water treatment equipment renewal: Install advanced water treatment equipment in drinking water treatment plants or high-risk water sources, especially for pollution that is difficult to treat through source reduction.

[0084] S4-2. Strategy simulation.

[0085] Reinforcement learning algorithms, such as Q-learning or deep reinforcement learning, are used to simulate and learn strategies. During the model training process, different strategy combinations are continuously tested and the strategy selection is adjusted based on the feedback of the reward function to find the optimal pollution reduction solution. At the same time, the knowledge of hydrological dynamics and pollutant behavior obtained from traditional process models is embedded in the learning process to improve the decision-making quality and adaptability of the model to the environment. After the model training is completed, the model is deployed in the actual area to monitor the effect of the pollution reduction strategy in real time and collect environmental data and socio-economic feedback after implementation. Based on these actual data, the model parameters and strategy selection are further optimized and adjusted to ensure that the model can continuously adapt to new environmental conditions and governance solution requirements.

[0086] Embodiment 2:

[0087] This example uses China as the target area to conduct an analysis on the control of perchlorate point source pollution.

[0088] Step 1: Environmental variable data collection.

[0089] Through sensor networks, remote sensing technology and literature research, raster data of environmental variables closely related to water pollution are systematically collected. These data cover climate information (such as temperature and precipitation), geographical environmental characteristics (such as topography and land use), hydrological data (such as river flow and water quality indicators), and detailed characteristics of industrial enterprises (such as location, industry type and pollutant emissions). Data sources include the National Meteorological Administration, environmental protection agencies, geographic information system (GIS) databases, and peer-reviewed scientific research articles to ensure the traceability and scientificity of the data sources. Strict data validation methods are used, including automated error detection and manual review, to identify and correct entry errors, fill in missing values, and handle outliers. In addition, the data is standardized to adapt to subsequent statistical analysis and model building. These strict data management measures ensure the accuracy and reliability of the analysis results.

[0090] Step 2: Development of water source pollution risk prediction model.

[0091] After data preprocessing, the relationship between environmental variables and pollution risk was modeled using a variety of machine learning algorithms, including support vector machines (SVM), random forests (RF), Catboost, and multilayer perceptrons (MLP). The effectiveness of the models was evaluated by 5-fold cross-validation and independent test sets, and performance indicators such as precision and recall were mainly used to measure the predictive ability of each model. The best performing model was subjected to hyperparameter adjustment using the Bayesian optimization method to ensure the optimization of parameter configuration. Subsequently, by applying these fine-tuned models for feature importance analysis, 11 key environmental variables (relevant enterprise density, annual average temperature, surface soil ocs, rainy season rainfall, groundwater level, deep soil BBOD, surface groundwater classification, deep soil cfvo, dry season rainfall, drought rate, dry season temperature) were identified. These variables are used to improve the resolution of predictions and mapping, so as to more accurately identify and display high-risk areas.

[0092] The results of the analysis indicated that surface water sources in a certain region were at high risk of exceeding the perchlorate limit. Based on this finding, the region was selected as a focus area for further research and development of pollution control strategies to reduce the risk of perchlorate contamination in drinking water sources in the region.

[0093] Step 3: Physical diffusion process simulation model.

[0094] SCHISM (Semi-implicit Cross-scale Hydroscience Integrated System Model) was used to simulate water flow conditions in the dry and rainy seasons. The initial setting of the model builds a three-dimensional hydrodynamic model of the river based on topographic data, and the input parameters include real-time water level, flow rate, and pollutant loads of major industrial outlets. In order to represent the concentration of different pollutants, different numbers of particles are put into the model, and the movement and distribution of these particles are used to simulate the diffusion path and concentration changes of pollutants in the water body.

[0095] Spatial data analysis was performed using ArcGIS, and the simulation results were processed using GIS technology to provide a detailed picture of the diffusion process of pollutants in the river system. Model parameter adjustments, including optimization of diffusion coefficient, sedimentation rate, and number of diffusing particles, were performed based on literature research and field monitoring data to ensure the accuracy and practicality of the simulation results.

[0096] Step 4: Pollution reduction optimization model.

[0097] Reinforcement learning algorithms were used to develop a pollution reduction optimization model, in which key environmental variables and hydrological variables, the former from preliminary machine learning analysis results and the latter from detailed hydrodynamic simulations, were used as input parameters. The designed reward function comprehensively considers reducing the number of affected people and reducing the frequency of pollution exceeding the standard, so as to motivate the model to find the best emission reduction strategy. The actions defined in the reinforcement learning model include different emission reduction strategies, such as industrial transformation and production process optimization. The changes in environmental status caused by the implementation of these strategies are again input into the water source pollution risk prediction model. The model re-evaluates the changed pollution risk and predicts the changes in the probability of perchlorate exceeding the standard in drinking water sources in various regions. In order to verify and calibrate the accuracy of the prediction results, a physical diffusion process model is further used, which simulates the actual diffusion path and concentration changes of pollutants in water bodies to ensure that the model's predictions are consistent with the actual situation. In this optimization process, a single agent, under the multiple goals of weighing economic development and drinking water safety, finds the best pollution reduction strategy that balances environmental protection and economic benefits through continuous iteration and adjustment.

[0098] Step 5: Comprehensive strategy formulation.

[0099] By iterating the reinforcement learning algorithm multiple times until the model performance converges, a comprehensive pollution reduction strategy is developed, which includes special management and monitoring of high-risk pollution areas and adjustments to specific emission standards for industrial enterprises.

[0100] The above description is only used to help understand the method of the present invention and its core essence, but the protection scope of the present invention is not limited thereto. For those skilled in the art in the art, equivalent replacement or change according to the technical solution and inventive concept of the present invention within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for controlling perchlorate point source pollution driven by environmental variables, characterized in that: The following steps are involved: S1, collect environmental variable data; Collect environmental variable data related to water source pollution in the target area; environmental variable data include pollution information data, climate data, geographical environment data, hydrological data and industrial enterprise characteristic data; S2. Identify high-risk pollution areas; Develop a water source pollution risk prediction model based on machine learning using environmental variable data, determine the key environmental variables for predicting water source pollutants exceeding the standard, and build a water source pollution risk prediction model based on the key environmental variables to predict the probability of water source pollutants exceeding the standard; use the water source pollution risk prediction model to identify the target area and obtain the high-risk pollution area within the target area; S201, pre-processing the environmental variable data, and classifying the pollution information data into two categories, namely, exceeding the standard and not exceeding the standard; S202, based on the preprocessed environmental variable data, respectively training the support vector machine, random forest, Catboost and multi-layer perceptron algorithms; S203, using precision, recall and F1 score as evaluation indicators, selecting the best performing model, and performing hyperparameter optimization on the selected model parameters through Bayesian optimization strategy; S204, using the model's built-in feature importance assessment method, the SHAP value-based method, and the statistical t-test method to identify key environmental variables that predict excessive water source pollutants; S205. Reconstruct a water source pollution risk prediction model based on key environmental variables and evaluate the water source pollution risk prediction model; S206. Predict the probability of water source pollutants exceeding the standard based on the water source pollution risk prediction model and identify high-risk pollution areas; S3. Calculate the pollution status of drinking water sources; A physical diffusion process simulation model is constructed based on hydrological variables to simulate the diffusion paths of pollutants in rivers in different seasons. The physical diffusion process simulation model is used to simulate the diffusion paths of point source pollutants in high-risk pollution areas, calculate the pollutant concentrations of drinking water sources in target areas, and dynamically simulate the pollution status of drinking water sources. The SCHISM model is used to simulate water flow conditions in the dry and rainy seasons and to construct a simulation model for the physical diffusion process; S4. Determine the optimal pollution reduction strategy; The key environmental variables in S2 and the hydrological variables in S3 are used as the environmental states of the pollution reduction optimization model, and the reward function is designed to build a pollution reduction optimization model driven by reinforcement learning; different pollution reduction optimization strategies are designed to motivate the pollution reduction optimization model to find the optimal pollution reduction strategy; Using reinforcement learning algorithm, key environmental variables and hydrological variables are used as environmental states of the pollution reduction optimization model, and the reward function of the pollution reduction optimization model is designed. The knowledge of hydrological dynamics and pollutant behavior obtained from the traditional process model is embedded in the learning process. Adjust the policy selection based on the feedback of the reward function to find the optimal pollution reduction strategy; The reward function uses the water source pollution risk prediction model as the calculation basis, calculates the probability of exceeding the standard for different drinking water sources, and defines the number of affected people based on the probability of exceeding the standard and the population distribution map. Ultimately, the number of affected people caused by the drinking water source pollution exceeding the standard is used as the reward benchmark; S5. Develop comprehensive pollution reduction strategies; Develop a comprehensive pollution reduction strategy based on the output of the pollution reduction optimization model; And according to the real-time monitoring data and the calculation results of the physical diffusion process simulation model, the pollution reduction strategy is adjusted to achieve dynamic pollution management.

2. The method for controlling perchlorate point source pollution driven by environmental variables according to claim 1, characterized in that: The pollution state of the drinking water source is calculated in S3, as follows: S301. Analyze high pollution risk areas using GIS software to determine the source of drinking water sources in the target area; S302, based on hydrological variables, the SCHISM model is used to simulate water flow conditions in the dry season and the rainy season, and a physical diffusion process simulation model is constructed; S303, calibrating a physical diffusion process simulation model using actual observation data in the target area; S304. The physical diffusion process simulation model performs physical diffusion simulation on point source pollutants discharged by industrial enterprises in high-risk pollution areas, and performs quantitative calculation of the pollution status of drinking water sources.

3. The method for controlling perchlorate point source pollution driven by environmental variables according to claim 2, characterized in that: Different emission reduction optimization strategies are designed in S4, as follows: The emission reduction optimization strategy includes industrial transformation and water treatment equipment renewal in high-risk pollution areas, and the implementation cost of the corresponding emission reduction optimization strategy is calculated and the strategy is adjusted accordingly; the industrial transformation of the industry is determined according to the pollution degree of point source pollutants to reduce the discharge of point source pollutants and reduce pollutants in drinking water sources; Water treatment equipment is updated at drinking water sources or high-risk pollution areas to purify and reduce pollutants in drinking water sources.

4. The method for controlling perchlorate point source pollution driven by environmental variables according to claim 3, characterized in that: The calculation results of the physical diffusion process simulation model in S5 are as follows: Based on the physical diffusion process simulation model, the diffusion paths of point source pollutants in high-risk pollution areas under different emission reduction optimization strategies are simulated to determine the pollutant concentrations in drinking water sources as a correction to the output results of the pollution reduction optimization model.

5. The perchlorate point source pollution control system driven by environmental variables obtained according to any one of the methods described in claims 1-4, characterized in that: include: Environmental variable data acquisition module, which integrates multiple sensors and data collection devices to collect climate, geographical environment, hydrology and industrial enterprise data, as well as drinking water source pollution information data; Water source pollution risk prediction model, which uses geographic environment and climate data as feature inputs, and uses drinking water source pollution information as the target variable to build a machine learning model to predict the probability distribution of drinking water source pollution and identify high-risk areas; Physical diffusion process simulation model, which establishes diffusion models for dry season and rainy season respectively, is used to calculate the pollutant concentration of drinking water sources to evaluate the impact of point source pollutants on drinking water sources; The pollution reduction optimization model integrates key environmental variables and hydrological variables as environmental states for reinforcement learning, implements knowledge-guided reinforcement learning, and uses the number of people affected by excessive drinking water source pollution as the main reward to determine the optimal pollution reduction strategy.

Images