Simulation Method, System, and Equipment for Agricultural Management Schemes Based on Improved Agent Model

Abstract

This invention provides a method, system, and equipment for simulating agricultural management measures based on an improved surrogate model, belonging to the field of agricultural non-point source pollution control technology. The method includes: acquiring various data from the study watershed and constructing a SWAT model for the watershed; calculating the monthly landscape yield of hydrological variables in each sub-watershed under different agricultural management measures; constructing corresponding improved response matrices based on the monthly landscape yields of hydrological variables in each sub-watershed under different agricultural management measures; using the improved response matrix method to calculate the outlet load of each sub-watershed; employing the improved response matrix method as a surrogate model; and using the surrogate model to optimize agricultural management measures and determine the optimal agricultural management measure scheme. This invention enables faster and more precise simulation and optimization of agricultural management measures, providing a reliable guarantee for the control of watershed non-point source pollution.

Description

Technical Field

[0001] This invention belongs to the field of agricultural non-point source pollution control technology, and relates to a method, system and equipment for simulating agricultural management measures based on an improved agent model. Background Technology

[0002] Currently, the SWAT (Soil and Water Assessment Tool) model, as a representative of distributed watershed hydrological models, divides the watershed into several sub-basins based on watershed elevation. Each sub-basin is further subdivided into different hydrological response units (HRUs) based on land use, soil type, and slope. The SWAT model can be used to simulate water quality and quantity, predict the impact of different agricultural management practices on the watershed's hydrology, sediment, and nitrogen and phosphorus nutrients, and is one of the more commonly used simulation models for watershed non-point source pollution control measures.

[0003] However, simulations based on physical process models are computationally complex, time-consuming, and computationally expensive. For example, simulating hydrological and nutrient loading using models like SWAT and MODFLOW can take anywhere from a few minutes to several hours. Especially when physical process models are coupled with various optimization models, the models need to run thousands of times (e.g., in searching for optimal management decisions), resulting in highly complex and time-consuming calculations. This leads to slow selection of agricultural management measures and difficulties in the optimization process. Therefore, surrogate models are needed to replace traditional watershed hydrological models.

[0004] Proxy modeling directly captures the relationship between the input and output of a physical mechanism model without considering the complex processes within the watershed system, thus reducing computational complexity and alleviating the problem of heavy workload. However, existing surrogate modeling techniques, such as the Receptance Matrix (RM) method and machine learning methods, essentially replace the original physical mechanism model with a simpler response relationship. This high efficiency is achieved at the expense of computational fidelity. In some sub-watersheds with strong human interference such as reservoir interception, the response relationship differs from other natural watersheds, and existing surrogate modeling methods are no longer effective in simulation. This, to some extent, limits the application of surrogate modeling in watershed pollution simulation research.

[0005] Therefore, it is urgent to improve the process and structure of the proxy model to balance the high fidelity and computational efficiency of the simulation model, thereby ensuring the accurate and rapid simulation and evaluation of watershed non-point source pollution control measures. Summary of the Invention

[0006] The purpose of this invention is to provide a response matrix proxy modeling method that considers improved strategies, which can be used to evaluate and decide on best management practices (BMPs) in agriculture to serve non-point source pollution control. The simulation method, system, and equipment for agricultural management practices based on the improved response matrix proxy model can simulate and optimize agricultural management practices more quickly and accurately, providing a reliable guarantee for the control of watershed non-point source pollution.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] A simulation method for agricultural management measures based on an improved response matrix surrogate model is disclosed. The method first acquires various data points for the study watershed and constructs a SWAT model of the watershed to calculate the monthly landscape yield of hydrological variables in each sub-watershed under different agricultural management measures. Second, based on the monthly landscape yield of hydrological variables in each sub-watershed under different agricultural management measures, an improved response matrix is ​​constructed, and the outlet load of each sub-watershed is initially calculated using the improved response matrix. Third, based on the initial load calculation results, for sub-watersheds with strong human activity disturbances such as reservoir interception, updated load calculation results are obtained using the intra-flow response formula and the reservoir interception formula. Finally, based on the updated load results, the optimal agricultural management measures are selected. The method includes the following steps:

[0009] Acquire various data from the study watershed and construct the spatial and attribute databases required for the operation of the SWAT model; the various data include DEM elevation, land use, soil type, meteorological and hydrological data of the study watershed;

[0010] The first step is to construct a SWAT model of the study watershed based on spatial and attribute databases; the study watershed includes multiple sub-watersheds and hydrological response units; specifically:

[0011] 1.1) Based on the spatial and attribute databases of the watershed, the study watershed is divided into several sub-watersheds and hydrological response units, and a preliminary watershed SWAT model is established.

[0012] 1.2) Use the global sensitivity analysis method (the SUFI-2 algorithm built into the model) to perform parameter sensitivity analysis and parameter calibration on the preliminary SWAT model of the watershed, and update and confirm the optimal parameter values ​​of the watershed SWAT model;

[0013] 1.3) The accuracy of the simulation results of the SWAT model with optimal parameter values ​​is evaluated by the Nash efficiency coefficient and the coefficient of determination. If both the Nash efficiency coefficient and the coefficient of determination are greater than 0.8, the SWAT model with updated parameters is considered to be completed.

[0014] The second step involves using the SWAT model of the watershed constructed in the first step to simulate the monthly landscape yield of hydrological variables in each sub-watershed under different agricultural management practices (BMPs), and to calculate the response matrix corresponding to each agricultural management practice; specifically:

[0015] 2.1) Using the SWAT model constructed for the research watershed, different agricultural management measures were simulated. During the simulation, each agricultural management measure was applied to all agricultural type hydrological response units in all sub-watersheds. The agricultural management measures included non-engineering measures and engineering measures. The non-engineering measures included different fertilization systems, irrigation strategies, and tillage methods. The engineering measures included vegetated filter strips and grassed waterways.

[0016] 2.2) Monthly landscape yields of hydrological variables in each sub-basin obtained from different agricultural management practices, denoted as y. m,t , representing the landscape yield in month t when the m-th agricultural management measure BMP is applied across 100% of the land area of ​​the entire watershed; the hydrological variables include: flow rate, total phosphorus, total nitrogen, and nitrate.

[0017] 2.3) Monthly landscape yield y of hydrological variables in each sub-basin under different agricultural management measures m,t It is the response matrix Y m,t The constituent elements, Y m,t It is a two-dimensional matrix listing the behavior month and the types of agricultural management measures, as shown in the instruction manual. Figure 2 As shown in (b).

[0018] The third step is to analyze the response matrix Y of different agricultural management practices (BMPs) obtained in the second step. m,t Preliminary calculation of the hydrological landscape yield load value Q at the outlet of each sub-basin after hydrological confluence. t Specifically:

[0019] 3.1) Construct a decision vector F representing the weights of agricultural management measures. m , represents the percentage of the area in which the m-th agricultural management measure BMP is applied in a certain sub-basin, with a range of 0-100%, where 100% means that the entire sub-basin is used;

[0020] 3.2) The total area A of each sub-basin is calculated based on the watershed spatial database, which is a vector of the areas of all sub-basins;

[0021] 3.3) Construct a connection matrix W describing the upstream and downstream relationships of each sub-basin to approximately characterize the hydrological basin processes in the SWAT model. Specifically, for each element w in the connection matrix... i,j If sub-watershed j is upstream of sub-watershed i, the off-diagonal element wi,j|i≠j =1, otherwise w i,j|i≠j =0; when i=j, it represents the diagonal element w of each sub-basin itself. i,j =1. As per the instruction manual. Figure 2 As shown in (d).

[0022] 3.4) Based on the response matrix Y m,t Decision vector F m Given the total area A and the connectivity matrix W, the outlet load of each sub-basin is calculated using the following formula:

[0023]

[0024] Where, q t ∈R N Let Y be the hydrological landscape yield load vector generated in the N sub-basins of month t; m,t ) Q ∈R N Let A be the response matrix of the m-th BMPs applied to 100% land area of ​​N sub-watersheds in month t; A∈R N F is a vector representing the areas of N sub-basins; m ∈R N Let be the vector representing the area proportion applied to the m-th BMP in the N sub-basins; diag(·) is the function that diagonalizes the vector; Q t ∈R N Let be the vector of hydrological landscape productivity load at the outlet of the Nth sub-basin in month t. Hydrological landscape productivity load includes flow rate, total nitrogen, total phosphorus, and nitrate nitrogen, etc.

[0025] The fourth step involves updating the load calculation results at the outlets of each sub-basin using the intra-flow reaction formula, based on the preliminary load calculation results obtained in the third step. Additionally, for sub-basins with strong human activity interference such as reservoir interception, the load calculation results are updated using the reservoir interception formula.

[0026] 4.1) Extract the pollutant-related terms (including the values ​​of total nitrogen, total phosphorus, and nitrate nitrogen) and the sub-basin flow values ​​from the outlet of each sub-basin in the SWAT model of the watershed (except for sub-basins with reservoirs, which are calculated independently according to Section 4.2). Establish the regression relationship between the pollutant-related terms and the flow, and obtain the regression coefficients. Then, calculate the updated value of the pollution load based on the intra-flow reaction formula (3). Add the updated value to the preliminary load calculation results of each sub-basin outlet obtained in the third step to obtain the updated load results. Taking total phosphorus (TP) as an example, the intra-flow reaction formula is as follows:

[0027]

[0028] Among them, ΔTP t ∈RN This is a vector representing the total phosphorus loss due to reaction within the river channel at the outlets of N sub-basins, obtained by summing a total phosphorus-related term and a flow-related term; TP_sum t ∈R N To generate the total phosphorus load for each sub-basin in month t, as initially calculated in step three; This is a vector containing the regression coefficients of total phosphorus production lost due to sedimentation in N sub-basins and total phosphorus production flowing into the river channel. This is a vector representing the total phosphorus loss due to algal absorption, decomposition, and benthic organism absorption, containing the reciprocal regression coefficients of total phosphorus and water yield across N sub-basins. The calculation process for the total phosphorus index above can be replaced by pollutant indices such as total nitrogen and nitrate nitrogen, thereby realizing the calculation process for the intra-flow reactions of different pollutants.

[0029] 4.2) Specifically, for sub-basins with strong human activity disturbances such as reservoir interception, the monthly inflow and outflow hydrological landscape yield loads of the sub-basin containing the reservoir are extracted from the SWAT model, and a linear regression is established. The regression coefficients are then used to update the preliminary load calculation results. Taking total phosphorus (TP) as an example, the calculation formula is as follows:

[0030]

[0031] in, and These are the inflow and outflow loads of the reservoir in the sub-basin in month t, obtained by SWAT model simulation of the baseline scenario; and These are the linear regression coefficients for inbound and outbound loads under the baseline scenario; (w' i,j ) TP To improve the total phosphorus linkage matrix considering the interception effect of reservoirs, the variables of each element are: D is the set of all downstream sub-basins of the reservoirs, and U is the set of all upstream sub-basins of the reservoirs; TP min ∈R N This is a vector representing the minimum total phosphorus load intercepted by the reservoir in each sub-basin, with the value being the value for the sub-basin immediately downstream of the reservoir. The value is 0 in all other sub-basins; TP t ∈R N The total phosphorus load (TP) for each sub-basin in month t, as initially calculated in step three, is generated. t,end ∈R N To take into account the updated total phosphorus load of each sub-basin in month t after reservoir interception.

[0032] The fifth step, based on the updated hydrological landscape yield load results (including total nitrogen, total phosphorus, nitrate nitrogen, etc.) obtained in the fourth step, uses these results as ecological scores for different watershed management practices (BMPs) at the watershed outlet. Simultaneously, the total cost of different BMPs is calculated as an evaluation criterion. The optimal spatial combination of watershed BMPs is selected by combining the ecological scores and the evaluation criteria, providing a scientific basis for watershed non-point source pollution control decisions. Specifically:

[0033] 5.1) The formulas for calculating the ecological scores and evaluation criteria for different agricultural management practices (BMPs) are as follows: f 1-3 Here is the formula for the ecological score, and f4 is the formula for the evaluation criteria:

[0034]

[0035] Where t is the calculation step size, in months; T is the total number of simulation periods, in months; TN outlet_avg TP outlet_avg and NO3-N outlet_avg These represent the multi-year average total nitrogen, total phosphorus, and nitrate nitrogen loads at the watershed outlet; TN mod_outlet,t TP mod_outlet,t and NO3-N mod_outlet,t Here, represents the updated total nitrogen, total phosphorus, and nitrate nitrogen load values ​​at the watershed outlet, calculated in step four, in kg; G represents the cost under different spatial allocation combinations of BMPs, in ten thousand yuan; for the study of N sub-watersheds and M BMP scenarios, A∈R N Let hm be the vector representing the agricultural land area of ​​each sub-basin. 2 ;F∈R N×M The spatial allocation matrix represents the proportion of the application area of ​​each sub-basin allocated to each BMPs scenario; C∈R M This is a vector representing the unit area cost corresponding to M BMPs scenarios applied to 100% of the land area of ​​the entire watershed, expressed in yuan / hm². 2 ,x represents the proportion of the application area of ​​different BMPs measures in each sub-basin, i.e., F in formula (1). m .

[0036] 5.2) Set the weights of the three ecological scores and evaluation criteria in the optimization objective to be the same. The combination of agricultural management measures with the highest final score is the optimal treatment decision for watershed non-point source pollution. The optimization objective formula is as follows:

[0037] F(x)=0.25f1(x)+0.25f2(x)+0.25f3(x)+0.25f4(x) (8)

[0038] In the formula, the letters represent the same meaning as in formula (7).

[0039] A simulation system for agricultural management measures based on an improved agent model includes:

[0040] The data acquisition module is used to acquire various types of data from the study watershed and to construct the spatial database and attribute database required for the operation of the SWAT model. The various types of data include DEM elevation, land use, soil type, meteorological and hydrological data of the study watershed.

[0041] The SWAT model building module is used to build a SWAT model of the study watershed based on a spatial database and an attribute database; the study watershed includes multiple sub-watersheds and hydrological response units;

[0042] The monthly landscape yield calculation module is used to simulate different agricultural management measures using the SWAT model and calculate the monthly landscape yield of hydrological variables in each sub-basin under different agricultural management measures.

[0043] The sub-basin outlet load calculation module is used to construct the corresponding response matrix based on the monthly landscape yield of hydrological variables of each sub-basin under different agricultural management measures, and to use the improved response matrix method to calculate the outlet load of each sub-basin.

[0044] An improved surrogate model determination module is used to employ an improved response matrix method as an improved surrogate model based on the surrogate calculation results of the outlet load of each sub-basin.

[0045] The optimal agricultural management measure scheme determination module is used to select the optimal agricultural management measure scheme by using an improved agent model.

[0046] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned agent-based agricultural management scheme simulation method.

[0047] Optionally, the memory is a non-transitory computer-readable storage medium.

[0048] The beneficial effects of this invention are as follows:

[0049] This invention provides a method, system, and equipment for simulating agricultural management measures based on an improved surrogate model. The method includes: constructing a SWAT model of the study watershed based on various data from a spatial and attribute database; calculating the monthly landscape yield of hydrological variables in each sub-watershed under different agricultural management measures using a surrogate model based on an improved response matrix method; further considering intra-flow response and reservoir interception effects; and finally determining the optimal agricultural management measures. This invention achieves high-fidelity and rapid simulation and optimization of agricultural management measures, overcoming the problems of high complexity, long processing time, and difficulty in nested optimization of SWAT models when simulating agricultural management measures for watershed non-point source pollution. It also overcomes the problem that previous surrogate modeling methods could not effectively simulate sub-watersheds with strong human interference such as reservoir interception by improving the response matrix. Thus, it balances the high fidelity of the simulation model with computational efficiency, providing a reliable guarantee for the accurate and rapid evaluation of optimal non-point source pollution control measures in watersheds. Attached Figure Description

[0050] Figure 1 A flowchart illustrating the simulation method for agricultural management measures based on an improved agent model provided by this invention;

[0051] Figure 2 A schematic diagram illustrating the agent model determination process provided by this invention;

[0052] Figure 3 A schematic diagram illustrating the principle of the improved response matrix method provided by this invention, taking into account intra-fluid reactions. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] The purpose of this invention is to provide a method, system, and device for simulating agricultural management measures based on an improved agent model, which can simulate and optimize agricultural management measures more quickly and accurately, providing a reliable guarantee for the control of watershed non-point source pollution.

[0055] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] See Figure 1 The present invention discloses a method for simulating agricultural management measures based on an improved agent model, comprising:

[0057] Step 1: Obtain various types of data from the research watershed and construct the spatial and attribute databases required for the SWAT model to run.

[0058] The various data sources include DEM elevation data, land use data, soil type data, meteorological data, and hydrological data for the study watershed. The spatial database for the study watershed is typically determined based on its natural geographical characteristics. This spatial database primarily includes data such as the digital elevation map (DEM), land use distribution map, soil type distribution map, and meteorological station distribution map. The attribute database mainly includes data on land use types, soil properties, meteorological element values, hydrological data, and water quality data for the study watershed. Meteorological element values ​​include rainfall, temperature, solar radiation, wind speed, and relative humidity.

[0059] In one specific embodiment, the SWAT model requires a large amount of input data during simulation calculations, mainly divided into two categories: spatial data and attribute data. Spatial data includes DEM, land use, soil type, slope, aspect, etc., while attribute data includes meteorological data such as daily relative humidity and daily evaporation, soil physicochemical properties, and hydrological data such as rainfall, sediment, and runoff from hydrological stations, as well as water quality data such as total nitrogen, total phosphorus, and nitrate during actual monitoring. This provides the data foundation for constructing the distributed SWAT model (hereinafter referred to as the SWAT model).

[0060] Step 2: Construct a SWAT model of the study watershed based on the spatial database and attribute database; the study watershed includes multiple sub-watersheds and hydrological response units.

[0061] Specifically, the SWAT model, combined with ArcGIS software, was used to extract the watershed system based on the DEM. The watershed was divided into multiple sub-watersheds by setting sub-watershed division thresholds and sub-watershed outlet points. Furthermore, plots within each sub-watershed with the same land use type, soil type, and slope value were grouped into a single hydrological response unit. Then, the global sensitivity analysis method of the SUFI-2 algorithm built into SWAT-CUP was used to perform parameter sensitivity analysis and parameter calibration on the SWAT model to determine the optimal parameter values. Further analysis was conducted using the Nash coefficient (NSE) and the coefficient of determination (R²). 2 The evaluation index is used to evaluate the accuracy of the simulation results of the SWAT model with the optimal parameter values, and the SWAT model with the higher evaluation result is regarded as the well constructed SWAT model.

[0062] Step 3: Use the SWAT model to simulate different agricultural management measures, preliminarily calculate the monthly landscape yield of hydrological variables in each sub-basin under different agricultural management measures, and calculate the response matrix corresponding to each agricultural management measure.

[0063] Step 3.1: Using the pre-constructed SWAT model for the research watershed, simulate the different agricultural management measures considered. These measures include non-engineering measures for source control such as different fertilization regimes, irrigation strategies, and tillage methods, as well as engineering measures such as vegetation filters and grassed waterways. The simulation of different agricultural management measures applies each management measure to all agricultural types (HRUs) in all sub-watersheds. The preliminary monthly landscape yields of hydrological variables for each sub-watershed, output by different agricultural management measures, are obtained. These hydrological variables include: flow, total phosphorus, total nitrogen, and nitrate.

[0064] Step 3.2: Use monthly-scale landscape yields of hydrological variables in each sub-basin under different agricultural management measures to construct the response matrix Y. m,t , making Y m,t A two-dimensional matrix listing the months of the behavior and the types of agricultural management measures, as shown in the instruction manual. Figure 2 As shown in (b).

[0065] Step 4: Response matrix Y for different agricultural management measures m,t Preliminary calculations were performed on the hydrological landscape output load values ​​at the outlets of each sub-basin after the hydrological confluence.

[0066] Step 4.1: Construct a decision vector F representing the weights assigned to agricultural management measures. m , representing the proportion of the m-th agricultural management measure BMP applied in a certain sub-basin;

[0067] Step 4.2: Calculate the total area A of each sub-basin based on the watershed spatial database;

[0068] Step 4.3: Construct a connectivity matrix W describing the upstream and downstream relationships of each sub-basin to approximately characterize the hydrological basin processes in the SWAT model. Specifically, for each element w in the connectivity matrix... i,j If sub-watershed j is upstream of sub-watershed i, the off-diagonal element w i,j|i≠j =1, otherwise w i,j|i≠j =0; when i=j, it represents the diagonal element w of each sub-basin itself. i,j =1. As per the instruction manual. Figure 2 As shown in (d).

[0069] Step 4.4: Based on the response matrix Y m,t Decision vector F m Given the total area A and the connectivity matrix W, calculate the outlet load of each sub-basin using formulas (1) and (2):

[0070]

[0071] Where, q t ∈R NLet Y be the hydrological landscape yield load vector generated in the N sub-basins of month t; m,t ) Q ∈R N Let A be the response matrix of the m-th BMPs applied to 100% land area of ​​N sub-watersheds in month t; A∈R N F is a vector representing the areas of N sub-basins; m ∈R N Let be the vector representing the area proportion applied to the m-th BMP in the N sub-basins; diag(·) is the function that diagonalizes the vector; Q t ∈R N Let be the vector of hydrological landscape productivity load at the outlet of the Nth sub-basin in month t. Hydrological landscape productivity load includes flow rate, total nitrogen, total phosphorus, and nitrate nitrogen, etc.

[0072] See the above steps. Figure 2 Based on the improved response matrix method, the hydrological landscape yield load values ​​at the outlet of each sub-basin after hydrological confluence were initially calculated:

[0073] (a): Using a well-constructed and calibrated watershed SWAT model, the agricultural management measures to be considered are set on HRUs of all agricultural types in all sub-watersheds. The resulting output load can be regarded as the load value when each measure is applied to 100% of the area of ​​each sub-watershed.

[0074] (b): The response matrix Y was obtained by simulating the monthly landscape yield of each sub-basin under various agricultural management measures. m,t The indicators include flow rate, total nitrogen, total phosphorus, and nitrate.

[0075] (c): Construct a decision vector F representing the proportion of sub-basin area allocated to agricultural management measure m. m The load generated by each measure is reduced by using the proportion of the decision vector.

[0076] (d): Construct a connection matrix W describing the upstream and downstream relationships, where the diagonal elements w i,i =1; if sub-basin j is upstream of sub-basin i, then the off-diagonal element w i,j|i≠j =1, otherwise w i,j|i≠j =0.

[0077] (e): Calculate the total area A of each sub-basin.

[0078] (f): Using formulas (9) and (10), the in-channel load of each sub-basin outlet of each indicator under the decision vector allocation scenario is initially calculated.

[0079] Step 5: Based on the preliminary load calculation results, update the load calculation results using the intra-flow reaction formula. In addition, for sub-basins with strong human activity interference such as reservoir interception, update the load calculation results using the reservoir interception formula.

[0080] Step 5.1: Extract the pollutant-related terms (including the values ​​of total nitrogen, total phosphorus, and nitrate nitrogen) and the sub-basin flow values ​​from the outlet of each sub-basin in the SWAT model of the watershed (except for sub-basins with reservoirs, which are calculated independently according to Section 4.2). Establish the regression relationship between the pollutant-related terms and the flow, and obtain the regression coefficients. Then, calculate the updated value of the pollution load based on the intra-flow reaction formula (3). Add the updated value to the preliminary load calculation results of each sub-basin outlet obtained in the third step to obtain the updated load results. Taking total phosphorus (TP) as an example, the intra-flow reaction formula is as follows:

[0081]

[0082] ΔTP t ∈R N This is a vector representing the total phosphorus loss due to reaction within the river channel at the outlets of N sub-basins, obtained by summing a total phosphorus-related term and a flow-related term; TP_sum t ∈R N To generate the total phosphorus load for each sub-basin in month t, as initially calculated in step three; This is a vector containing the regression coefficients of total phosphorus production lost due to sedimentation in N sub-basins and total phosphorus production flowing into the river channel. This is a vector representing the total phosphorus loss due to algal absorption, decomposition, and benthic organism absorption, containing the reciprocal regression coefficients of total phosphorus and water yield across N sub-basins. The calculation process for the total phosphorus index above can be replaced by pollutant indices such as total nitrogen and nitrate nitrogen, thereby realizing the calculation process for the intra-flow reactions of different pollutants.

[0083] Step 5:2: Specifically, for sub-basins with strong human activity disturbances such as reservoir interception, the monthly inflow and outflow hydrological landscape yield loads of the sub-basin containing the reservoir are extracted from the SWAT model, and a linear regression is established. The regression coefficients are then used to update the preliminary load calculation results. Taking total phosphorus (TP) as an example, the calculation formula is as follows:

[0084]

[0085] in, and These are the inflow and outflow loads of the reservoir in the sub-basin in month t, obtained by SWAT model simulation of the baseline scenario; and These are the linear regression coefficients for inbound and outbound loads under the baseline scenario; (w' i,j )TP To improve the total phosphorus linkage matrix considering the interception effect of reservoirs, the variables of each element are: D is the set of all downstream sub-basins of the reservoirs, and U is the set of all upstream sub-basins of the reservoirs; TP min ∈R N This is a vector representing the minimum total phosphorus load intercepted by the reservoir in each sub-basin, with the value being the value for the sub-basin immediately downstream of the reservoir. The value is 0 in all other sub-basins; TP t ∈R N The total phosphorus load (TP) for each sub-basin in month t, as initially calculated in step three, is generated. t,end ∈R N To take into account the updated total phosphorus load of each sub-basin in month t after reservoir interception.

[0086] Appendix 3 shows a schematic diagram illustrating the principle of reactions within the total phosphorus stream, characterizing the processes of phosphorus deposition, mineralization, exchange with algae, and release from benthic organisms, which can be represented as follows:

[0087] (a): First-order reaction ΔP=k P ·P,k P This is the phosphorus loss coefficient.

[0088] (b): Zero-order reaction ΔP = k, which depends on flow depth or algal concentration. P ·f(depth) or ΔP=k P ·f(algae).

[0089] Meanwhile, the amount of phosphorus transformed or exchanged during transport is also related to the residence time in the river channel, as shown in formulas (9) and (10):

[0090] ΔP=k P,P ·ΔT·P (9)

[0091] ΔP=k P,Q ·f(depth,algae)·ΔT (10)

[0092] In the formula, ΔP represents the amount of phosphorus transformed or exchanged during transport, and k P,P k represents the phosphorus loss coefficient in a zero-order reaction. P,Q The coefficient represents the phosphorus loss in the first-order reaction, ΔT represents the residence time of phosphorus in the channel, and P represents the amount of phosphorus in the channel. It is worth noting that the SWAT model assumes that phosphorus and sediment transport are completely decoupled once they reach the river network. Although sediment loss in the landscape has a significant impact on phosphorus loss in the landscape in both reality and SWAT, sediment deposition and erosion have no effect on phosphorus transport in SWAT. Since the goal of this invention is to surrogate the SWAT model, sediment is not considered in the phosphorus treatment process of in-flow reactions.

[0093] Step 6: Using the updated hydrological landscape yield load results (including total nitrogen, total phosphorus, nitrate nitrogen, etc.), these are considered as ecological scores for different agricultural management measures at the watershed outlet. Simultaneously, the total cost of different agricultural management measures is calculated as the evaluation criterion. The optimal spatial combination of watershed agricultural management measures is then selected by combining the ecological scores and the evaluation criteria.

[0094] Step 6.1: The formulas for calculating the ecological scores and evaluation criteria for different agricultural management practices (BMPs) are as follows: f 1-3 Here is the formula for the ecological score, and f4 is the formula for the evaluation criteria:

[0095]

[0096] Where t is the calculation step size, in months; T is the total number of simulation periods, in months; TN outlet_avg TP outlet_avg and NO3-N outlet_avg These represent the multi-year average total nitrogen, total phosphorus, and nitrate nitrogen loads at the watershed outlet; TN mod_outlet,t TP mod_outlet,t and NO3-N mod_outlet,t Here, represents the updated total nitrogen, total phosphorus, and nitrate nitrogen load values ​​at the watershed outlet, calculated in step four, in kg; G represents the cost under different spatial allocation combinations of BMPs, in ten thousand yuan; for the study of N sub-watersheds and M BMP scenarios, A∈R N Let hm be the vector representing the agricultural land area of ​​each sub-basin. 2 ;F∈R N×M The spatial allocation matrix represents the proportion of the application area of ​​each sub-basin allocated to each BMPs scenario; C∈R M This is a vector representing the unit area cost corresponding to M BMPs scenarios applied to 100% of the land area of ​​the entire watershed, expressed in yuan / hm². 2 ,x represents the proportion of the application area of ​​different BMPs measures in each sub-basin, i.e., F in formula (1). m .

[0097] Step 6.2: Set the weights of the three ecological scores and evaluation criteria in the optimization objective to be the same. Finally, calculate the combination of agricultural management measures with the highest score as the optimal treatment decision for watershed non-point source pollution. The optimization objective formula is as follows:

[0098] F(x)=0.25f1(x)+0.25f2(x)+0.25f3(x)+0.25f4(x) (8)

[0099] The specific implementation process of step 6 is as follows:

[0100] (a): Simulate all the agricultural management measures set and select the measures with better non-point source reduction effect. In this embodiment, five typical measures are selected.

[0101] (b): Selected typical measures are combined and allocated in different ways in one (or more) sub-basins to form different agricultural management measures.

[0102] In this embodiment, the total number of schemes for combining and allocating the five typical measures is [number missing]. Species, among which This represents the number of possible solutions when x measures are selected from five typical measures, and it is assumed that the measures selected from the five typical measures distribute the area of ​​this sub-basin evenly. This is the decision vector of the x selected measures. The decision vector F of the 5-x unselected measures m =0.

[0103] (c): Calculate the ecological score and evaluation criteria of each scheme after different combinations of allocation, and obtain the scheme with the best reduction effect and lowest cost of each nutrient index as the optimal agricultural management scheme.

[0104] Applying the optimal agricultural management measures selected in this invention to the prevention and control of watershed non-point source pollution can greatly improve the effectiveness of watershed non-point source pollution control and treatment.

[0105] This invention provides a method, system, and equipment for simulating agricultural management measures based on an improved surrogate model. Taking advantage of the independent calculation of HRU units in the SWAT watershed hydrological model and the ability to employ different measures in different units, this invention achieves faster and more precise simulation and optimization of agricultural management measures through data acquisition, construction of a distributed SWAT hydrological model, surrogate application of the SWAT model using an improved response matrix method, consideration of improved processes involving intra-flow reactions and reservoir interception, and optimization of agricultural management scenarios. This provides a reliable guarantee for the control of watershed non-point source pollution. Compared to existing technologies, this invention not only overcomes the challenges of high complexity, long time consumption, and difficulty in optimization in simulating agricultural management scenarios for watershed non-point source pollution using the SWAT hydrological model, but also significantly improves the efficiency of watershed non-point source pollution control measures by utilizing the surrogate model concept. Furthermore, compared to existing surrogate models, it considers processes such as pollutant reactions in the river channel and reservoir interception, achieving a more accurate surrogate application of the SWAT model. This is of great significance for using model-based control technologies to manage agricultural non-point source pollution.

[0106] Based on the method provided by this invention, this invention also provides a simulation system for agricultural management measures based on an improved agent model, comprising:

[0107] The data acquisition module is used to acquire various types of data from the study watershed and to construct the spatial database and attribute database required for the operation of the SWAT model. The various types of data include DEM elevation, land use, soil type, meteorological and hydrological data of the study watershed.

[0108] The SWAT model building module is used to build a SWAT model of the study watershed based on a spatial database and an attribute database; the study watershed includes multiple sub-watersheds and hydrological response units.

[0109] The monthly landscape yield calculation module is used to simulate different agricultural management measures using the SWAT model and calculate the monthly landscape yield of hydrological variables in each sub-basin under different agricultural management measures.

[0110] The sub-basin outlet load calculation module is used to construct the corresponding response matrix based on the monthly landscape yield of hydrological variables of each sub-basin under different agricultural management measures, and to use the improved response matrix method to calculate the outlet load of each sub-basin.

[0111] The surrogate model determination module is used to employ an improved response matrix method as the surrogate model based on the surrogate calculation results of the outlet load of each sub-basin.

[0112] The module for determining the optimal agricultural management measures is used to select the best agricultural management measures by employing a proxy model.

[0113] Furthermore, the present invention also provides an electronic device, which may include a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The processor can call a computer program stored in the memory to execute the aforementioned agent-based agricultural management simulation method.

[0114] Furthermore, when the computer program in the aforementioned memory is implemented as a software functional unit and sold or used as an independent product, it can be stored in a non-transitory computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0115] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0116] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this 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 this invention.

Claims

1. A simulation method for agricultural management measures based on an improved agent model, characterized in that, The aforementioned agricultural management measure simulation method includes the following steps: The first step is to acquire various types of data for the study watershed and construct a SWAT model for the study watershed based on spatial and attribute databases; the study watershed includes multiple sub-watersheds and hydrological response units. The second step involves using the SWAT model of the watershed constructed in the first step to simulate the monthly landscape yield of hydrological variables in each sub-watershed under different agricultural management practices (BMPs), and to calculate the response matrix corresponding to each agricultural management practice. The third step is to analyze the response matrices of different agricultural management practices (BMPs) obtained in the second step. Preliminary calculations of the hydrological landscape yield load values ​​at the outlets of each sub-basin after hydrological confluence. ; The fourth step involves updating the load calculation results at the outlets of each sub-basin using the intra-flow reaction formula, based on the preliminary load calculation results obtained in the third step. For sub-basins with strong human activity interference due to reservoir interception, the updated load calculation results are obtained using the reservoir interception formula. The details are as follows: 4.1) First, extract the pollutant-related terms and sub-basin flow values ​​at the outlet of each sub-basin in the SWAT model of the watershed, establish the regression relationship between the pollutant-related terms and flow, and obtain the regression coefficients. The pollutant-related terms include the values ​​of total nitrogen, total phosphorus and nitrate nitrogen. Then, calculate the updated value of the pollution load based on the intra-flow reaction formula (3), and add the updated value to the preliminary load calculation results of each sub-basin outlet obtained in the third step to obtain the updated load results. Using total phosphorus (TP) as an example, the intra-flow reaction formula is as follows: ； in, This is a vector representing the total phosphorus loss due to reaction within the river channel at the outlets of N sub-basins, obtained by summing a total phosphorus-related term and a flow-related term. To generate the total phosphorus load for each sub-basin in month t, as initially calculated in step three; This is a vector containing the regression coefficients of total phosphorus production lost due to sedimentation in N sub-basins and total phosphorus production flowing into the river channel. This is a vector containing the total phosphorus loss due to algal absorption, decomposition, and benthic organism absorption, along with the reciprocal regression coefficients of water yield across N sub-basins. The calculation process for total phosphorus can be replaced by total nitrogen, nitrate nitrogen, or other pollutant indicators to realize the calculation process for the intra-stream reaction of different pollutants; 4.2) For the disturbed sub-basins, extract the monthly inflow and outflow hydrological landscape production loads of the sub-basins where the reservoir is located from the SWAT model, establish a linear regression, and obtain its regression coefficients to update the preliminary load calculation results. The formula for calculating total phosphorus (TP) is as follows: ； in, and These are the inflow and outflow loads of the reservoir in the sub-basin in month t, obtained by SWAT model simulation of the baseline scenario; and These are the linear regression coefficients of the inbound and outbound loads under the baseline scenario, respectively. To improve the total phosphorus linkage matrix to take into account the interception effect of reservoirs, the variables of each element are: D is the set of all downstream sub-basins of the reservoirs, and U is the set of all upstream sub-basins of the reservoirs. This is a vector representing the minimum total phosphorus load intercepted by the reservoir in each sub-basin, with the value being the value for the sub-basin immediately downstream of the reservoir. The value is 0 in all other sub-basins; To generate the total phosphorus load for each sub-basin in month t, as initially calculated in step three; To consider the updated total phosphorus load of each sub-basin in month t after reservoir interception; The fifth step involves using the updated hydrological landscape yield load results obtained in the fourth step as the ecological score of different agricultural management measures (BMPs) at the watershed outlet. At the same time, the total cost of different agricultural management measures (BMPs) is calculated as the evaluation criterion. The optimal spatial combination of watershed agricultural management measures (BMPs) is selected by combining the ecological score and the evaluation criterion, providing a basis for watershed non-point source pollution control decisions.

2. The method for simulating agricultural management measures based on an improved agent model according to claim 1, characterized in that, The first step is specifically as follows: 1.1) Based on the spatial and attribute databases of the watershed, the study watershed is divided into several sub-watersheds and hydrological response units, and a preliminary watershed SWAT model is established; 1.2) Use global sensitivity analysis to perform parameter sensitivity analysis and parameter calibration on the preliminary SWAT model of the watershed, and update and confirm the optimal parameter values ​​of the watershed SWAT model; 1.3) The accuracy of the simulation results of the SWAT model with optimal parameter values ​​is evaluated by the Nash efficiency coefficient and the coefficient of determination. If both the Nash efficiency coefficient and the coefficient of determination are greater than 0.8, the SWAT model with updated parameters is considered to be completed.

3. The method for simulating agricultural management measures based on an improved agent model according to claim 2, characterized in that, The second step is specifically as follows: 2.1) Using the SWAT model constructed for the research watershed, simulate the different agricultural management measures to be considered. During the simulation, each agricultural management measure is applied to all agricultural type hydrological response units in all sub-watersheds. 2.2) Monthly landscape yields of hydrological variables in each sub-basin obtained from different agricultural management practices. Landscape yields will be denoted as follows: , representing the landscape yield in month t when the m-th agricultural management measure BMP is applied across 100% of the land area of ​​the entire watershed; the hydrological variables include: flow rate, total phosphorus, total nitrogen, and nitrate; 2.3) Monthly landscape yields of hydrological variables in each sub-basin under different agricultural management measures It is the response matrix The constituent elements, It is a two-dimensional matrix of behavioral months and categories of agricultural management measures.

4. The method for simulating agricultural management measures based on an improved agent model according to claim 3, characterized in that, In 2.1), the agricultural management measures include non-engineering measures and engineering measures; the non-engineering measures include different fertilization systems, irrigation strategies and tillage methods; the engineering measures include vegetation filter strips and grassed waterways.

5. The method for simulating agricultural management measures based on an improved agent model according to claim 3, characterized in that, The third step is specifically as follows: 3.1) Construct a decision vector representing the weights of agricultural management measures. , represents the percentage of the area in which the m-th agricultural management measure BMP is applied in a certain sub-basin, with a range of 0-100%, where 100% means that the entire sub-basin is used; 3.2) The total area A of each sub-basin is calculated based on the watershed spatial database, which is a vector of the areas of all sub-basins; 3.3) Construct a connection matrix describing the upstream and downstream relationships of each sub-basin. This is used to approximately characterize the hydrological watershed processes in the SWAT model; where, for each element in the connection matrix... If sub-watershed j is upstream of sub-watershed i, non-diagonal elements =1, otherwise =0; when i=j, it represents the diagonal element of each sub-basin itself. =1; 3.4) Based on the response matrix Decision vector Given the total area A and the connectivity matrix W, the outlet load of each sub-basin is calculated using the following formula: ； in, Let be the hydrological landscape output load vector generated in the N sub-basins in month t; Let m be the response matrix of the m-th BMPs applied to 100% land area of ​​N sub-watersheds in month t; A vector representing the areas of N sub-basins; Let m be the vector representing the area ratio applied to the m-th BMP across the N sub-basins; This is a function that diagonalizes a vector; Let be the vector of hydrological landscape productivity load at the outlet of the Nth sub-basin in month t, where hydrological landscape productivity load includes flow, total nitrogen, total phosphorus and nitrate nitrogen.

6. The method for simulating agricultural management measures based on an improved agent model according to claim 1, characterized in that, The fifth step is specifically as follows: 5.1) The calculation formulas for the ecological scores and evaluation criteria of different agricultural management practices (BMPs) are as follows: This is the formula for ecological score. The evaluation criteria formula is as follows: ； Where t is the calculation step size, in months; T represents the total number of simulation periods, in months; , and These are the multi-year average total nitrogen, total phosphorus, and nitrate nitrogen load values ​​at the watershed outlet; , and These represent the updated total nitrogen, total phosphorus, and nitrate nitrogen load values ​​at the watershed outlet, calculated in step four, in kg; G represents the cost under different spatial allocation combinations of BMPs, in ten thousand yuan; for the study of N sub-watersheds and M BMP scenarios, Let hm be the vector representing the agricultural land area of ​​each sub-basin. 2 ; The spatial allocation matrix represents the proportion of the application area of ​​each sub-basin allocated to each BMPs scenario. This is a vector representing the unit area cost corresponding to M BMPs scenarios applied to 100% of the land area of ​​the entire watershed, expressed in yuan / hm². 2 , x represents the proportion of the application area of ​​different BMPs measures in each sub-basin, i.e., in formula (1). ; 5.2) Set the weights of the three ecological scores and evaluation criteria in the preferred objective to be equal. Calculate the combination of agricultural management measures with the highest score as the optimal treatment decision for watershed non-point source pollution. The objective formula is as follows: ； In the formula, the letters represent the same as in formula (7).

7. A simulation system for agricultural management measures based on an improved agent model, employing the simulation method for agricultural management measures based on an improved agent model as described in claim 1, characterized in that, include: The data acquisition module is used to acquire various types of data from the study watershed and to construct the spatial database and attribute database required for the operation of the SWAT model. The various types of data include DEM elevation, land use, soil type, meteorological and hydrological data of the study watershed. The SWAT model building module is used to build a SWAT model of the study watershed based on a spatial database and an attribute database; the study watershed includes multiple sub-watersheds and hydrological response units; The monthly landscape yield calculation module is used to simulate different agricultural management measures using the SWAT model and calculate the monthly landscape yield of hydrological variables in each sub-basin under different agricultural management measures. The sub-basin outlet load calculation module is used to construct the corresponding response matrix based on the monthly landscape yield of hydrological variables of each sub-basin under different agricultural management measures, and to use the improved response matrix method to calculate the outlet load of each sub-basin. An improved surrogate model determination module is used to employ an improved response matrix method as an improved surrogate model based on the surrogate calculation results of the outlet load of each sub-basin. The optimal agricultural management measure scheme determination module is used to select the optimal agricultural management measure scheme by using an improved agent model.

8. An electronic device that runs a simulation method for agricultural management measures based on an improved agent model according to claims 1-6, characterized in that, The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the simulation method for agricultural management measures based on the improved agent model.

9. An electronic device according to claim 8, characterized in that, The memory is a non-transitory computer-readable storage medium.

Images