Decision support system and method for agricultural non-point source pollution control based on hybrid modeling

Abstract

This invention relates to a decision support system and method for agricultural non-point source pollution control based on hybrid modeling, belonging to the field of agricultural environmental management and intelligent decision-making technology. This invention integrates multi-source agricultural-related data to construct multi-type databases. The core is a random forest intelligent prediction model trained using agricultural hydrological model simulation data combined with actual statistical data, nested within the model library. This model can accurately calculate agricultural non-point source pollution loads and support multi-scenario simulations. The system can achieve full-chain decision support from pollution assessment to measure optimization, quantitatively evaluate the effectiveness of pollution prevention and control measures, and provide a scientific basis for the formulation of control plans. This invention effectively solves the problems of data fragmentation, poor model applicability, and lack of systematic decision-making in traditional methods, significantly improving the precision and management efficiency of regional agricultural non-point source pollution control.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of smart agriculture and environmental pollution control, specifically relating to a decision support system and method for agricultural non-point source pollution control based on hybrid modeling. Background Technology

[0002] Agricultural non-point source pollution, especially nitrogen and phosphorus entering water bodies through runoff and leaching due to excessive application of agricultural fertilizers, has become a major source of water pollution in my country and globally. Its emissions are dispersed, highly random, and its mechanisms of action are complex, making its remediation far more difficult than that of point source pollution. Traditional remediation methods mainly rely on human intervention to determine management strategies, resulting in low efficiency and poor sustainability.

[0003] Currently, the main challenge in non-point source pollution control lies in how to scientifically quantify the environmental pollution risks under different management measures and find the optimal balance between agricultural productivity (irrigation water productivity, nitrogen fertilizer partial productivity, and crop yield) and environmental effects (pollution load). Existing technologies have significant shortcomings: 1. Complex mechanistic models: Although hydrological or biogeochemical models such as SWAT and DNDC can simulate pollution processes, they have many parameters, are difficult to calibrate, and have high computational costs, making it difficult to promote and apply them on a large scale in actual production.

[0004] 2. Lack of intelligent decision support: Most studies only focus on pollution potential assessment or yield prediction, failing to develop a complete technical solution that integrates prediction and optimization, and thus unable to provide farmers or policymakers with specific and actionable strategies for "high yield and environmental protection".

[0005] 3. Single objective: Existing optimization methods often pursue only a single objective, such as maximizing yield or minimizing fertilizer application, making it difficult to reconcile the real-world conflicts between multiple objectives in agricultural production, such as high yield, high efficiency, and low pollution.

[0006] Therefore, there is an urgent need in this field for an intelligent method and system that can quickly and accurately assess the nitrogen pollution load of farmland and automatically generate the optimal water and nitrogen management strategy that takes into account both production and environmental protection objectives. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the purpose of this invention is to comprehensively utilize technologies such as multi-source databases, Web visualization methods, random forest algorithms, and VIKOR multi-criteria decision-making to quantitatively calculate and analyze irrigation water productivity, nitrogen fertilizer partial productivity, crop yield, and nitrogen pollution load under different irrigation nitrogen application scenarios in different hydrological years in a region. Ultimately, it aims to determine a suitable irrigation nitrogen application strategy that balances farmland productivity improvement with non-point source pollution control, thereby achieving intelligent, efficient, and refined management of agricultural water and nitrogen in irrigation areas.

[0008] To achieve the above objectives, in a first aspect, the present invention provides a decision support system for agricultural non-point source pollution control based on hybrid modeling, including an identity authentication module, a data layer module, a model layer module, a decision optimization module, a visualization module, a knowledge base module, and an external resource integration module, wherein the modules are interconnected. The identity authentication module is used to receive the identity credential information input by the user and send it to the backend authentication service for verification, and control the user's system access permissions based on the verification result; The data layer module includes an attribute database and a spatial database, built on a MySQL database. It is used to collect and store multi-source data such as meteorological, agricultural, water quality, soil and socio-economic indicators to form a regional agricultural hydrological database. It also stores a fertilizer composition database. The data layer module receives data query requests from clients and parses pagination parameters to complete data queries and return results. The model layer module includes a random forest prediction model, a pollution load calculation model, and a VIKOR multi-criteria decision model. The random forest prediction model is trained using regional agricultural hydrological databases and AHC model simulation data. The pollution load calculation model is constructed based on the empirical coefficient method and can calculate the pollution load of typical agricultural non-point source pollutants such as nitrogen and phosphorus. The decision optimization module first receives crop planting parameters and environmental conditions, simulates yields under various irrigation nitrogen reduction scenarios based on a trained random forest model, and calculates irrigation water productivity and nitrogen fertilizer partial productivity; determines the loss coefficient based on regional nitrogen loss, calculates nitrogen pollution load using a built-in nitrogen pollution load rapid assessment model; and uses the VIKOR multi-criteria decision method to comprehensively evaluate indicators such as crop yield, irrigation water productivity, nitrogen fertilizer partial productivity, and nitrogen pollution load, and automatically recommends the optimal irrigation nitrogen management plan. The visualization module is used to receive parameters such as crops and geographical location input by users and output pollution load calculation results, displaying the spatial distribution map of the potential reduction of nitrogen application for irrigation of major crops under different hydrological year types in the study area and the simulation results of typical case studies. The knowledge base module is used to integrate and display multi-source environmental standards such as surface water environmental quality standards and groundwater quality standards, as well as various production and discharge coefficient manuals, and to provide query services for laws, regulations, policies and technical standards. The external resource integration module is used to categorize and display links to authoritative websites of government departments, research institutions, industry organizations, and other organizations related to agricultural non-point source pollution control decisions, enabling one-stop information access.

[0009] Secondly, the present invention provides a decision support method for agricultural non-point source pollution control based on hybrid modeling, applied to the agricultural non-point source pollution control decision support system based on hybrid modeling described in the first aspect, comprising the following steps: S1. System Access and Authentication: S1.1 The system's identity authentication module provides a user login interface, receives the user's input identity credential information, and sends it to the backend authentication service for verification. The identity authentication module controls user access permissions based on the verification result: if the verification is successful, the user is redirected to the system's main operation interface to use the decision support function.

[0010] S2. Agricultural hydrological data collection and database establishment and retrieval: S2.1. Through the data layer module of the system, combined with regional field measurement data and statistical data such as statistical yearbooks, multi-source data such as meteorological, agricultural, water quality, soil and socio-economic indicators are collected, and table structures are set and data is imported in the MySQL database of the data layer module to form a regional agricultural hydrological database. S2.2 Data processing and access steps: The data layer module of the system receives the data query request from the client, parses the pagination parameters, queries the database based on the parameters, obtains the corresponding data results, and returns them to the client; S3. Train a random forest model using AHC simulation results and statistical data: S3.1 Input the regional agricultural hydrological database data collected in the data layer module of AHC, establish AHC models for different crops, use the calibrated and verified AHC models to simulate crop yield under the preset irrigation nitrogen application scenario, and calculate irrigation water productivity, nitrogen fertilizer partial productivity and nitrogen pollution load to obtain AHC model simulation data. S3.2. The obtained AHC model simulation data and regional agricultural hydrological database statistical data are used as the dataset. The dataset is divided into training set and validation set. The random forest model is trained through the model layer module to ensure that the model accuracy is within a reasonable range. S4. Intelligent Decision-Making Optimization Steps: Input crop planting parameters and environmental conditions, call the trained random forest model to simulate yield under various irrigation nitrogen reduction scenarios, calculate irrigation water productivity and nitrogen fertilizer partial productivity, use the built-in nitrogen pollution load rapid assessment model, determine the loss coefficient in combination with regional nitrogen loss, calculate the nitrogen pollution load under each irrigation nitrogen reduction scenario, and then call the VIKOR multi-criteria decision method from the model layer module to comprehensively evaluate indicators such as crop yield, irrigation water productivity, nitrogen fertilizer partial productivity and nitrogen pollution load. The decision optimization module automatically recommends the optimal irrigation nitrogen management plan. The VIKOR decision-making method is deployed in the code to establish a corresponding multi-objective decision-making program, and the irrigation amount, nitrogen application amount and corresponding decision variable data of the irrigation nitrogen application strategy are output through the visualization module. S5. Case Demonstration Steps: The system's visualization module displays the spatial distribution map of the reduction potential of irrigation water and nitrogen application for major crops under different hydrological year types in the study area, and provides a quantitative analysis of the reduction degree in combination with regional characteristics, providing a visual decision-making basis for the precise implementation of agricultural non-point source pollution control measures. S6. Pollution load calculation steps: The model layer module inputs crop information, geographical location, study area area, fertilizer application data and environmental parameters into the pollution load calculation model. Based on the fertilizer composition database information of the data layer module, it automatically estimates and outputs the total nitrogen and phosphorus pollution load in agricultural planting sources in the study area, providing a basis for precise pollution prevention and control measures and targeted treatment of pollution sources. S7. Knowledge Base Integration and Management Steps: The knowledge base module of the system integrates and displays multi-source environmental standards and production and discharge coefficient manuals, including surface water environmental quality standards, groundwater quality standards, soil environmental quality standards, livestock and poultry breeding pollutant discharge standards, and various production and discharge coefficient manuals, providing users with a unified platform for querying laws and policies and technical standards, and providing standardized basis for accurate calculation and scientific decision-making of agricultural non-point source pollution. S8. External resource integration step: The external resource integration module of the system integrates and categorizes links to authoritative websites of multiple fields related to agricultural non-point source pollution control decisions, including government departments, research institutions and industry organizations, to provide users with a one-stop channel for policy inquiry, technical reference and industry dynamics, and to expand the information sources and reference dimensions of the decision support system.

[0011] Furthermore, the core formula for calculating the random forest model in step S3 is shown in (1): ; In the formula: Y sim It is the final predicted output of the random forest model for a given input x; T It is the total number of decision trees in the random forest; f t (x) is the first t The prediction output of a decision tree for an input feature x; x is the input feature vector, i.e., the amount of irrigation water. x 1. Nitrogen application rate x 2. Precipitation x 3. Temperature x 4. Soil clay content x 5. Soil silt content x 6. Soil sand content x 7. Soil organic matter content x 8; Furthermore, the core algorithm of the nitrogen pollution load rapid assessment model built into step S4 is as shown in equation (2), where the nitrogen pollution load calculated is the nitrogen pollution load per unit area: ; In the formula: N load It is the nitrogen pollution load, kg·hm - ²; f load It is the nitrogen loss coefficient, which is determined based on regional measured data and literature; N application This refers to the nitrogen application rate under specific conditions, i.e., the fertilization intensity, expressed in kg·hm². - ².

[0012] Furthermore, the core algorithm of the VIKOR decision model in step S4 is as follows: ; ; ; In the formula: x ij For the first i Under the first irrigation nitrogen application scenario, the first j The value of each decision indicator; A j + For the first j The optimal solution for each decision metric across all scenarios; A j - For the first j The worst solution for each decision indicator across all scenarios; S i For the context i Distance to the optimal solution; R i For the context i Distance to the worst solution; w j For the first j The weight of each decision indicator; v The decision coefficient represents the decision-maker's preference; Q i The decision benefit ratio for each scenario is given, with the minimum value corresponding to the optimal scenario. Furthermore, in step S4, the decision coefficients v It is 0.5.

[0013] Furthermore, in step S4, the weights of the decision variables—crop yield, irrigation water productivity, nitrogen fertilizer partial productivity, and nitrogen pollution load—are determined comprehensively based on the agricultural production characteristics, soil hydrological conditions, and environmental management needs of the study area.

[0014] Furthermore, in step S4, when making decisions using a multi-objective decision-making procedure, crop yield, irrigation water productivity, and nitrogen fertilizer partial productivity are set as positive decision variables, while nitrogen pollution load is set as a negative decision variable.

[0015] Furthermore, the core formula of the pollution load calculation model in step S6 is shown in (6), which calculates the total pollution load of a certain type of pollutant in the study area: ; In the formula: L The total pollution load of a certain type of pollutant in the study area is expressed in kg·a. -1 , ; A area Let hm be the area of ​​the study area. 2 ; E output This represents the output emission coefficient of this pollutant in the study area; F input The fertilization intensity for this study area is expressed in kg·hm². - ²·a -1 .

[0016] Compared with other existing technologies, the present invention has the following advantages: 1. This invention overcomes the technical limitations of traditional irrigation nitrogen application decision-making methods in terms of multi-source heterogeneous data acquisition, multi-dimensional index quantification, and agricultural non-point source pollution decision-making and control. For the first time, it realizes the synergistic optimization of three-dimensional objectives of water security, food security, and ecological security, and provides a complete quantitative decision-making framework for multi-objective management of complex agricultural systems.

[0017] 2. By integrating machine learning prediction models with multi-criteria decision-making methods, this invention significantly reduces the time cost and subjective uncertainty of traditional methods, and realizes full automation from data collection and scenario simulation to scheme selection, providing a replicable and scalable technical paradigm for the precise management of water and nitrogen in the study area.

[0018] 3. Compared with traditional multi-objective decision-making tools, this invention provides an integrated interactive interface. Through one-click decision-making process design and intuitive result visualization, it significantly reduces the learning threshold for users, while enhancing the transparency of the decision-making process and the interpretability of the results, thus promoting the effective transformation of scientific research results into practical applications.

[0019] 4. By setting up two pollution load calculation models, this invention not only meets the needs of accurate nitrogen and phosphorus load calculation in scenarios such as environmental monitoring, but also avoids redundant calculations in the decision-making process, thereby improving system operating efficiency and scenario adaptability. Attached Figure Description

[0020] Figure 1 The flowchart shows the decision support method for agricultural non-point source pollution control based on hybrid modeling of the present invention. Figure 2 This is a display of the homepage service content of the decision support system for agricultural non-point source pollution control in the Hetao Irrigation District, as described in this embodiment of the invention. Figure 3 This is a display of attribute data for the decision support system for agricultural non-point source pollution control in the Hetao Irrigation District, as described in this embodiment of the invention. Figure 4 This is a spatial data display of the decision support system for agricultural non-point source pollution control in the Hetao Irrigation District, as described in this embodiment of the invention. Figure 5 This is a user-friendly interface showing the start and output results of the decision support system for agricultural non-point source pollution control based on hybrid modeling in this embodiment of the invention. Figure 6 This invention provides examples of scenarios for reducing irrigation and nitrogen application rates under different hydrological year types and current irrigation and fertilization conditions. (In this description, we have designed scenarios for reducing irrigation and nitrogen application rates for spring wheat and summer maize in the Hetao Irrigation District. Data for different irrigation and nitrogen application scenarios for other crop types are also applicable.) Figure 7 The decision index data obtained from simulations under different irrigation nitrogen application scenarios in different hydrological years in this embodiment of the invention (in this description, for example, crop yield, irrigation water productivity, nitrogen fertilizer partial productivity, and nitrogen pollution load of summer maize in 2019 and spring wheat in 2020 in the Hetao Irrigation District were simulated under different irrigation and nitrogen application scenarios. Combined with the statistical data of each township in 2021 as the basic data, a random forest model was trained, and the trained model was used to simulate the preset scenarios of each township. The simulation results were used to make multi-criteria decisions. Data on different irrigation nitrogen application scenarios for other crop types in different years can also be simulated). Figure 8 This is a spatial distribution map of the decision-making process for optimal irrigation nitrogen application measures in various townships under different irrigation nitrogen application scenarios in different hydrological years, according to an embodiment of the present invention. Figure 9 This is a display of the pollution load calculation interface of the decision support system for agricultural non-point source pollution control in the Hetao Irrigation District, as shown in this embodiment of the invention. Figure 10 This is a knowledge base demonstration of the decision support system for agricultural non-point source pollution control in the Hetao Irrigation District, as described in this embodiment of the invention. Figure 11 This is a display of the relevant website access pages for the decision support system for agricultural non-point source pollution control in the Hetao Irrigation District, as described in this embodiment of the invention. Detailed Implementation

[0021] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the following embodiments are given for illustrative purposes only and are not intended to limit the scope of the present invention. Those skilled in the art can make various modifications and substitutions to the present invention without departing from its spirit and essence.

[0022] The flowchart of the decision support method for agricultural non-point source pollution control based on hybrid modeling of the present invention is as follows: Figure 1 As shown. The specific implementation method is as follows: S1. Open the user login interface and enter your identity credentials. After successful verification, you will be redirected to the main system interface to use the decision support function.

[0023] S2. Agricultural hydrological data collection and database establishment and retrieval: Combining field measurement data and statistical data from yearbooks in the Hetao Irrigation District, data on meteorology, agriculture, water quality, soil, and socioeconomic indicators were collected. Table structures were set up and data imported into a MySQL database to form a regional agricultural hydrological database, which is then accessed and displayed on the front-end interface. The homepage showcases the various data and module distributions within the system (e.g., ...). Figure 2 As shown), the various data types collected from the Hetao Irrigation District are divided into attribute databases (such as...). Figure 3 (as shown) and spatial databases (such as) Figure 4 (As shown).

[0024] S3. Train a random forest model using simulation results and statistical data from the agricultural hydrological model AHC: S3.1 Input the regional agricultural hydrological database data collected in the data layer module of AHC, establish AHC models for different crops, use the calibrated and verified AHC models to simulate crop yield under the preset irrigation nitrogen application scenario, and calculate irrigation water productivity, nitrogen fertilizer partial productivity and nitrogen pollution load to obtain AHC model simulation data. S3.2. The obtained AHC model simulation data and regional agricultural hydrological database statistical data are used as the dataset. The dataset is divided into training set and validation set. The random forest model is trained through the model layer module to ensure that the model accuracy is within a reasonable range. The calculation method of the random forest model, its core formula is shown in (1): ; In the formula: Y sim It is the final predicted output of the random forest model for a given input x; T It is the total number of decision trees in the random forest; f t (x) is the first tThe prediction output of a decision tree for an input feature x; x is the input feature vector, i.e., the amount of irrigation water. x 1. Nitrogen application rate x 2. Precipitation x 3. Temperature x 4. Soil clay content x 5. Soil silt content x 6. Soil sand content x 7. Soil organic matter content x 8.

[0025] S4. Intelligent Decision-Making Optimization Steps: Inputting crop planting parameters and environmental conditions, a trained random forest model simulates yields under various irrigation nitrogen reduction scenarios (such as...). Figure 5 As shown), it calculates irrigation water productivity and nitrogen fertilizer partial productivity, uses a built-in rapid nitrogen pollution load assessment model, and combines it with the localized nitrogen loss coefficient to calculate the nitrogen pollution load under various irrigation nitrogen reduction scenarios. Then, it uses the VIKOR multi-criteria decision-making method to comprehensively evaluate indicators such as crop yield, irrigation water productivity, nitrogen fertilizer partial productivity, and nitrogen pollution load, and automatically recommends the optimal irrigation nitrogen management plan (e.g., Figure 5 (as shown) The core algorithm of the nitrogen pollution load rapid assessment model built into step S4 is shown in equation (2). The nitrogen pollution load calculated here is the nitrogen pollution load per unit area. ; In the formula: N load It is the nitrogen pollution load, kg·hm - ²; f load It is the nitrogen loss coefficient, which is determined based on regional measured data and literature; N application This refers to the nitrogen application rate under specific conditions, i.e., the fertilization intensity, expressed in kg·hm². - ².

[0026] The decision coefficient of the VIKOR multi-criteria decision-making method is 0.5 (compromise method), and the weight of each decision variable is determined according to the actual situation of the study area and the needs of the managers.

[0027] The core algorithm of the VIKOR decision model in step S4 is as follows: ; ; ; In the formula: x ij For the first i Under the first irrigation nitrogen application scenario, the firstj The value of each decision indicator; A j + For the first j The optimal solution for each decision metric across all scenarios; A j - For the first j The worst solution for each decision indicator across all scenarios; S i For the context i Distance to the optimal solution; R i For the context i Distance to the worst solution; w j For the first j The weight of each decision indicator; v The decision coefficient represents the decision-maker's preference; Q i The decision benefit ratio for each scenario is given, with the minimum value corresponding to the optimal scenario. In the above decision-making process, crop yield, irrigation water productivity, and nitrogen fertilizer partial productivity are set as positive decision variables, while nitrogen pollution load is set as a negative decision variable.

[0028] S5. Case Presentation Steps: The front-end interface displays different irrigation nitrogen application scenarios for major crops under different hydrological year types in the study area (such as...). Figure 6 As shown), decision indicators obtained through random forest model simulation, such as yield, irrigation water productivity, nitrogen fertilizer partial productivity, and nitrogen pollution load, are used. Figure 7 As shown), and spatial distribution maps of irrigation water and nitrogen application reduction potential after multi-criteria decision-making (as shown). Figure 8 As shown in the figure, it provides a visual basis for precise policy implementation in agricultural non-point source pollution control.

[0029] S6. Pollution load calculation steps: The pollution load calculation model in the model layer module is input with crop information, geographical location, study area area, fertilizer application data, and environmental parameters. Based on the fertilizer composition database information in the data layer module, the model automatically estimates and outputs the nitrogen and phosphorus pollution loads in agricultural non-point sources (e.g., ...). Figure 9 (As shown), this provides a basis for precise pollution prevention and control measures and targeted treatment of pollution sources; The core formula of the pollution load calculation model in step S6 is shown in (6), which calculates the total pollution load of a certain type of pollutant in the study area: ; In the formula: L The total pollution load of a certain type of pollutant in the study area is expressed in kg·a.-1 , ; A area Let hm be the area of ​​the study area. 2 ; E output This represents the output emission coefficient of this pollutant in the study area; F input The fertilization intensity for this study area is expressed in kg·hm². - ²·a -1 .

[0030] S7. Knowledge Base Integration and Management Steps: By querying multi-source environmental standards and pollution discharge coefficient manuals through the front-end interface, including surface water environmental quality standards, groundwater quality standards, soil environmental quality standards, livestock and poultry breeding pollutant emission standards, and various pollution discharge coefficient manuals, one can access a unified platform for querying regulations and policies and referencing technical standards (such as...). Figure 10 (As shown).

[0031] S8. External resource integration steps: The front-end interface provides access to links to websites of authoritative organizations across multiple fields related to agricultural non-point source pollution control decisions, including government departments, research institutions, and industry organizations, enabling one-stop access to policy inquiries, technical references, and industry updates (such as...). Figure 11 (As shown).

[0032] At each step, you can select "Exit" in the upper right corner of the system to log out and return to the login page. You will need to log in again the next time you enter the system.

[0033] This invention uses specific experimental data to illustrate the principles and implementation methods of the invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of the invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the invention.

Claims

1. A decision support system for agricultural non-point source pollution control based on hybrid modeling, characterized in that, It includes an identity authentication module, a data layer module, a model layer module, a decision optimization module, a visualization module, a knowledge base module, and an external resource integration module, with each module communicating with each other; The identity authentication module is used to receive the identity credential information input by the user and send it to the backend authentication service for verification, and control the user's system access permissions based on the verification result; The data layer module includes an attribute database and a spatial database, built on a MySQL database. It is used to collect and store multi-source data such as meteorological, agricultural, water quality, soil and socio-economic indicators to form a regional agricultural hydrological database. It also stores a fertilizer composition database. The data layer module receives data query requests from clients and parses pagination parameters to complete data queries and return results. The model layer module includes a random forest prediction model, a pollution load calculation model, and a VIKOR multi-criteria decision model. The random forest prediction model is trained using regional agricultural hydrological databases and AHC model simulation data. The pollution load calculation model is constructed based on the empirical coefficient method and can calculate the pollution load of typical agricultural non-point source pollutants such as nitrogen and phosphorus. The decision optimization module first receives crop planting parameters and environmental conditions, simulates yields under various irrigation nitrogen reduction scenarios based on a trained random forest model, and calculates irrigation water productivity and nitrogen fertilizer partial productivity; determines the loss coefficient based on regional nitrogen loss, calculates nitrogen pollution load using a built-in nitrogen pollution load rapid assessment model; and uses the VIKOR multi-criteria decision method to comprehensively evaluate indicators such as crop yield, irrigation water productivity, nitrogen fertilizer partial productivity, and nitrogen pollution load, and automatically recommends the optimal irrigation nitrogen management plan. The visualization module is used to receive parameters such as crops and geographical location input by users and output pollution load calculation results, displaying the spatial distribution map of the potential reduction of nitrogen application for irrigation of major crops under different hydrological year types in the study area and the simulation results of typical case studies. The knowledge base module is used to integrate and display multi-source environmental standards such as surface water environmental quality standards and groundwater quality standards, as well as various production and discharge coefficient manuals, and to provide query services for laws, regulations, policies and technical standards. The external resource integration module is used to categorize and display links to authoritative websites of government departments, research institutions, industry organizations, and other organizations related to agricultural non-point source pollution control decisions, enabling one-stop information access.

2. A decision support method for agricultural non-point source pollution control based on hybrid modeling, applied to the decision support system for agricultural non-point source pollution control based on hybrid modeling as described in claim 1, characterized in that, Includes the following steps: S1. System Access and Authentication: S1.1 The system's identity authentication module provides a user login interface, receives the user's input identity credential information, and sends it to the backend authentication service for verification; the identity authentication module controls user access permissions based on the verification result: if the verification is successful, it redirects to the system's main operation interface to use the decision support function; S2. Agricultural hydrological data collection and database establishment and retrieval: S2.

1. Through the data layer module of the system, combined with regional field measurement data and statistical data such as statistical yearbooks, multi-source data such as meteorological, agricultural, water quality, soil and socio-economic indicators are collected, and table structures are set and data is imported in the MySQL database of the data layer module to form a regional agricultural hydrological database. S2.2 Data processing and access steps: The data layer module of the system receives the data query request from the client, parses the pagination parameters, queries the database based on the parameters, obtains the corresponding data results, and returns them to the client; S3. Train a random forest model using AHC simulation results and statistical data: S3.1 Input the regional agricultural hydrological database data collected in the data layer module of AHC, establish AHC models for different crops, use the calibrated and verified AHC models to simulate crop yield under the preset irrigation nitrogen application scenario, and calculate irrigation water productivity, nitrogen fertilizer partial productivity and nitrogen pollution load to obtain AHC model simulation data. S3.

2. The obtained AHC model simulation data and regional agricultural hydrological database statistical data are used as the dataset. The dataset is divided into training set and validation set. The random forest model is trained through the model layer module to ensure that the model accuracy is within a reasonable range. S4. Intelligent Decision-Making Optimization Steps: Input crop planting parameters and environmental conditions, call the trained random forest model to simulate yield under various irrigation nitrogen reduction scenarios, calculate irrigation water productivity and nitrogen fertilizer partial productivity, use the built-in nitrogen pollution load rapid assessment model, determine the loss coefficient in combination with regional nitrogen loss, calculate the nitrogen pollution load under each irrigation nitrogen reduction scenario, and then call the VIKOR multi-criteria decision method from the model layer module to comprehensively evaluate indicators such as crop yield, irrigation water productivity, nitrogen fertilizer partial productivity and nitrogen pollution load. The decision optimization module automatically recommends the optimal irrigation nitrogen management plan. The VIKOR decision-making method is deployed in the code to establish a corresponding multi-objective decision-making program, and the irrigation amount, nitrogen application amount and corresponding decision variable data of the irrigation nitrogen application strategy are output through the visualization module. S5. Case Demonstration Steps: The system's visualization module displays the spatial distribution map of the reduction potential of irrigation water and nitrogen application for major crops under different hydrological year types in the study area, and provides a quantitative analysis of the reduction degree in combination with regional characteristics, providing a visual decision-making basis for the precise implementation of agricultural non-point source pollution control measures. S6. Pollution load calculation steps: The model layer module inputs crop information, geographical location, study area area, fertilizer application data and environmental parameters into the pollution load calculation model. Based on the fertilizer composition database information of the data layer module, it automatically estimates and outputs the total nitrogen and phosphorus pollution load in agricultural planting sources in the study area, providing a basis for precise pollution prevention and control measures and targeted treatment of pollution sources. S7. Knowledge Base Integration and Management Steps: The knowledge base module of the system integrates and displays multi-source environmental standards and production and discharge coefficient manuals, including surface water environmental quality standards, groundwater quality standards, soil environmental quality standards, livestock and poultry breeding pollutant discharge standards, and various production and discharge coefficient manuals, providing users with a unified platform for querying laws and policies and technical standards, and providing standardized basis for accurate calculation and scientific decision-making of agricultural non-point source pollution. S8. External resource integration step: The external resource integration module of the system integrates and categorizes links to authoritative websites of multiple fields related to agricultural non-point source pollution control decisions, including government departments, research institutions and industry organizations, to provide users with a one-stop channel for policy inquiry, technical reference and industry dynamics, and to expand the information sources and reference dimensions of the decision support system.

3. The decision support method for agricultural non-point source pollution control based on hybrid modeling according to claim 2, characterized in that, The calculation method of the random forest model in step S3 is shown in the core formula (1): ； In the formula: Y sim It is the final predicted output of the random forest model for a given input x; T It is the total number of decision trees in the random forest; f t (x) is the first t The prediction output of a decision tree for an input feature x; x is the input feature vector, i.e., the amount of irrigation water. x 1. Nitrogen application rate x 2. Precipitation x 3. Temperature x 4. Soil clay content x 5. Soil silt content x 6. Soil sand content x 7. Soil organic matter content x 8.

4. The decision support method for agricultural non-point source pollution control based on hybrid modeling according to claim 2, characterized in that, The core algorithm of the nitrogen pollution load rapid assessment model built into step S4 is shown in equation (2). The nitrogen pollution load calculated here is the nitrogen pollution load per unit area. ； In the formula: N load It is the nitrogen pollution load, kg·hm - ²; f load It is the nitrogen loss coefficient, which is determined based on regional measured data and literature; N application This refers to the nitrogen application rate under specific conditions, i.e., the fertilization intensity, expressed in kg·hm². - ².

5. The decision support method for agricultural non-point source pollution control based on hybrid modeling according to claim 2, characterized in that, The core algorithm of the VIKOR decision model in step S4 is as follows: ； ； ； In the formula: x ij For the first i Under the first irrigation nitrogen application scenario, the first j The value of each decision indicator; A j + For the first j The optimal solution for each decision metric across all scenarios; A j - For the first j The worst solution for each decision indicator across all scenarios; S i For the context i Distance to the optimal solution; R i For the context i Distance to the worst solution; w j For the first j The weight of each decision indicator; v The decision coefficient represents the decision-maker's preference; Q i The decision benefit ratio for each scenario is given, with the minimum value corresponding to the optimal scenario.

6. The decision support method for agricultural non-point source pollution control based on hybrid modeling according to claim 5, characterized in that, In step S4, the decision coefficient v It is 0.

5.

7. The decision support method for agricultural non-point source pollution control based on hybrid modeling according to claim 2, characterized in that, In step S4, the weights of the decision variables—crop yield, irrigation water productivity, nitrogen fertilizer partial productivity, and nitrogen pollution load—are determined comprehensively based on the agricultural production characteristics, soil hydrological conditions, and environmental management needs of the study area.

8. The decision support method for agricultural non-point source pollution control based on hybrid modeling according to claim 2, characterized in that, In step S4, when making decisions using a multi-objective decision-making procedure, crop yield, irrigation water productivity, and nitrogen fertilizer partial productivity are set as positive decision variables, while nitrogen pollution load is set as a negative decision variable.

9. The decision support method for agricultural non-point source pollution control based on hybrid modeling according to claim 2, characterized in that, The core formula of the pollution load calculation model in step S6 is shown in (6), which calculates the total pollution load of a certain type of pollutant in the study area: ； In the formula: L The total pollution load of a certain type of pollutant in the study area is expressed in kg·a. -1 , ; A area Let hm be the area of ​​the study area. 2 ; E output This represents the output emission coefficient of this pollutant in the study area; F input The fertilization intensity for this study area is expressed in kg·hm². - ²·a -1 .

Images