Agricultural manure soil crop interaction optimization regulation method and system

Abstract

The application provides an agricultural manure soil crop interaction optimization regulation method and system, divides a farmland unit for a farmland area, constructs a farmland unit intelligence and constructs a farmland unit social network containing a physical proximity network, a feature similarity network and a causal correlation network, the farmland unit intelligence agent carries out social learning based on the farmland unit social network, acquires individual performance data of manure regulation strategy, generates a candidate manure regulation strategy set, carries out strategy radiation simulation on the candidate manure regulation strategy set through the causal correlation network and the infectious disease model, acquires strategy radiation degree data, the global farmland unit intelligence agent carries out global simulation simulation based on the candidate manure regulation strategy set, acquires strategy global influence index, optimizes the candidate manure regulation strategy set according to the individual performance data, the strategy radiation degree data and the strategy global influence index, generates an agricultural manure regulation strategy, so as to improve the overall utilization rate of manure and the stability of farmland ecology.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agriculture, in particular to an agricultural manure soil crop interaction optimization regulation method and system. BACKGROUND

[0002] With the development of agricultural scale and precision, manure as an important source of organic nutrients, through the regulation of manure can achieve the effect of improving crop yield and protecting soil ecology. The traditional method of manure regulation in farmland mainly focuses on individual decision optimization and global effect evaluation, but it still has certain limitations, and it is often difficult to meet the local and global needs of farmland.

[0003] Specifically, there are causal relationship between farmland units, such as hydrological connection and pest transmission. Although the manure regulation strategy developed for a single farmland unit can significantly solve the local problem of the farmland unit, due to the causal relationship between farmland units, the local strategy for a farmland unit will inevitably have a positive or negative impact on its neighbor unit. The traditional method often ignores these impacts, making it difficult to balance the overall benefits of farmland when solving local problems of farmland units, resulting in low overall utilization of manure and difficulty in ensuring the stability of farmland ecology. SUMMARY

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the present application provides an agricultural manure soil crop interaction optimization regulation method, which comprises:

[0005] S1: dividing the farmland area into farmland units and constructing farmland unit agents;

[0006] S2: constructing a farmland unit social network for the farmland unit agent, which includes a physical proximity network, a feature similarity network and a causal relationship network;

[0007] S3: the farmland unit agent performs social learning based on the farmland unit social network, obtains individual performance data of the manure regulation strategy, and generates a candidate manure regulation strategy set;

[0008] S4: performing strategy radiation simulation on the candidate manure regulation strategy set through the causal relationship network and the infectious disease model to obtain strategy radiation data;

[0009] S5: performing global simulation of the farmland unit agent based on the candidate manure regulation strategy set to obtain a strategy global influence index;

[0010] S6: optimizing the candidate manure regulation strategy set according to the individual performance data, the strategy radiation data and the strategy global influence index to generate an agricultural manure regulation strategy.

[0011] As a further scheme of the present application, a farmland unit social network is constructed for the farmland unit agent, and the farmland unit social network includes a physical proximity network, a feature similarity network, and a causal correlation network, and includes:

[0012] The geometric center point of the farmland unit corresponding to the farmland unit agent is taken as a Venn diagram seed point, a physical proximity network is constructed based on the Venn diagram seed point, and the corresponding Venn diagram seed point is linked according to a pre-set edge connection rule, and a distance attenuation function is used to obtain the edge connection weight of the corresponding edge;

[0013] The farmland unit agent is taken as a feature similarity network node, feature extraction is performed on the farmland unit data set corresponding to each node through a multi-head attention mechanism and a deep residual network structure to generate a farmland unit feature vector, the feature similarity between the farmland unit feature vectors corresponding to each node is obtained, and an edge connection of the feature similarity network is established according to a pre-set feature similarity threshold, and the nodes with the edge connection between the nodes are defined as feature similarity nodes;

[0014] The farmland unit agent is taken as a causal correlation network node, and a full connection network is established for the causal correlation network node, a conditional independence test is performed on the full connection network, the edges determined as conditionally independent are removed, the causal correlation strength corresponding to each edge connection in the causal correlation network is synchronously obtained, and the causal correlation network is generated.

[0015] As a further scheme of the present application, the farmland unit agent performs social learning based on the farmland unit social network, obtains individual performance data of the manure regulation strategy, and generates a candidate manure regulation strategy set, including:

[0016] The farmland unit agent to be regulated by manure is defined as a target farmland unit agent, the spatial proximity nodes and the feature similarity nodes of the target farmland unit agent in the farmland unit social network are obtained, a proximity similar unit set is constructed, and the corresponding correlation data set is screened in combination with the causal correlation network, and a neighbor farmland unit agent set is constructed based on the proximity similar unit set and the correlation data set;

[0017] The target farmland unit agent performs historical optimal manure scheduling strategy learning on the neighbor farmland unit agent to obtain an initial candidate manure scheduling strategy set;

[0018] The target farmland unit agent performs counterfactual questioning on the neighbor farmland unit agent based on the initial candidate manure scheduling strategy set, the neighbor farmland unit agent calls the local causal correlation network to generate a counterfactual estimation result and sends it to the target farmland unit agent;

[0019] The counterfactual estimation results are evaluated based on the adversarial evaluation mechanism to obtain the evaluation results of the initial candidate manure scheduling strategies and generate a set of candidate manure scheduling strategies.

[0020] As a further aspect of the present invention, the counterfactual estimation results are evaluated based on an adversarial evaluation mechanism to obtain initial candidate manure scheduling strategy evaluation results, and a candidate manure scheduling strategy set is generated, including:

[0021] Obtain historical performance data for all farmland unit agents in the farmland unit social network. The historical performance data represents the deviation between the actual performance and the expected performance of the historical manure scheduling strategy. Select a predetermined number of farmland unit agents from the highest and lowest ranked historical performance data respectively to generate a candidate adversarial evaluation agent set.

[0022] Obtain the middle centrality of farmland unit agents within the candidate adversarial evaluation agent set in the farmland unit social network, select farmland unit agents within a predetermined threshold range, and generate an adversarial evaluation agent set;

[0023] The counterfactual estimation results are input into the set of adversarial evaluation agents for prediction and evaluation, generating initial candidate manure scheduling strategy evaluation results. Based on the initial candidate manure scheduling strategy evaluation results, a set of candidate manure scheduling strategies and corresponding individual performance data are generated, whereby the individual performance data represents the expected performance of the candidate manure scheduling strategies.

[0024] As a further aspect of the present invention, strategy radiation simulation is performed on the candidate manure regulation strategy set using causal association networks and infectious disease models to obtain strategy radiation data, including:

[0025] The diffusion process of the candidate manure regulation strategy set in the corresponding ecosystem of the whole farmland unit was simulated based on the causal relationship network and infectious disease model, and the strategy radiation data of each candidate manure regulation strategy in the candidate manure regulation strategy set were obtained.

[0026] The strategy radiation data represents the degree of radiation and spread of the candidate manure regulation strategy within a certain farmland unit to the entire farmland area.

[0027] As a further aspect of the present invention, the method further includes:

[0028] Based on the infectious disease model, nodes in the causal association network are divided into the infection period, the incubation period, and the immunity period;

[0029] The infection period represents that the candidate manure regulation strategy is adopted and implemented in the farmland unit, the incubation period represents that the candidate manure regulation strategy is adopted but not implemented in the farmland unit, and the immunity period represents that the candidate manure regulation strategy does not achieve the expected effect in the farmland unit and the farmland unit does not receive the candidate manure regulation strategy in the preset immune period;

[0030] Randomly select a plurality of causally related network nodes as infected nodes in the infection period in the infectious disease model, set the transmission rate as the conversion probability to perform transmission infection on the uninfected nodes, so that the uninfected nodes enter the incubation period, and define the nodes entering the incubation period as latent nodes, set the uninfected nodes to convert into infected nodes in the infection period with the lesion rate as the conversion probability, and set the infected nodes to convert into immune nodes in the immunity period with the immunity rate as the conversion probability;

[0031] Based on the Monte Carlo algorithm, the strategy radiation simulation is performed to generate the strategy radiation data of the farmland unit corresponding to the causally related network nodes.

[0032] As a further scheme of the present application, the global farmland unit agent performs global simulation and simulation based on the candidate manure regulation strategy set to obtain a strategy global influence index, including:

[0033] The global farmland unit agent matches the corresponding candidate manure regulation strategy from the candidate manure regulation strategy set, and obtains the strategy global influence index through simulation and simulation, and the strategy global influence index at least includes an absorption efficiency index, an environmental impact index and an economic benefit index, and the strategy global influence index is obtained based on the strategy global influence index;

[0034] The absorption efficiency index is used to quantify the absorption efficiency of soil nutrients in the global farmland unit after the global farmland unit agent executes the corresponding candidate manure regulation strategy;

[0035] The environmental impact index is used to quantify the influence rate of the global farmland on the environment after the global farmland unit agent executes the corresponding candidate manure regulation strategy;

[0036] The economic benefit index is used to quantify the fluctuation of the expected income of the global farmland crops after the global farmland unit agent executes the corresponding candidate manure regulation strategy.

[0037] As a further scheme of the present application, the candidate manure regulation strategy set is optimized according to the individual performance data, the strategy radiation data and the strategy global influence index to generate an agricultural manure regulation strategy, including:

[0038] Based on the strategy radiation data, the individual performance data of each candidate manure regulation strategy in the candidate manure regulation strategy set and the strategy global influence index of the candidate manure regulation strategy set are coordinated to generate an agricultural manure regulation strategy.

[0039] As a further scheme of the present application, a farmland region is divided into farmland units, and a farmland unit agent is constructed, comprising:

[0040] The continuous farmland region is divided into multiple farmland units of consistent size based on a geographic information system and a clustering algorithm, multi-source data of the multiple farmland units is collected, and the collected raw data is subjected to outlier rejection, missing value filling and data standardization to obtain a farmland unit dataset corresponding to each farmland unit, the farmland unit dataset at least including farmland soil data, crop growth data, weather data and farmland management data;

[0041] The farmland unit agent is constructed based on the farmland unit dataset and preset social learning rules and strategy triggering rules, wherein the social learning rules are used for social learning of the farmland unit and its adjacent farmland units similar in physical space or farmland characteristics, and the strategy triggering rules are used for triggering generation of a manure regulation strategy when the farmland unit reaches a predetermined condition.

[0042] In another aspect, the embodiment of the present application further provides an agricultural manure soil crop interaction optimization regulation system, comprising:

[0043] A unit division module, which is used for dividing a farmland region into farmland units and constructing a farmland unit agent;

[0044] A network construction module, which is used for constructing a farmland unit social network including a physical proximity network, a characteristic similarity network and a causal correlation network for the farmland unit agent;

[0045] A social learning module, which is used for social learning of the farmland unit agent, obtaining individual performance data of a manure regulation strategy, and generating a candidate manure regulation strategy set;

[0046] A radiation simulation module, which is used for strategy radiation simulation of the candidate manure regulation strategy set according to the causal correlation network and an infectious disease model, and generating strategy radiation degree data;

[0047] A global simulation module, which is used for global simulation of the global farmland unit agent according to the candidate manure regulation strategy set, and obtaining a strategy global influence index;

[0048] A strategy generation module, which is used for optimization of the candidate manure regulation strategy set according to the individual performance data, the strategy radiation degree data and the strategy global influence index, and generating an agricultural manure regulation strategy.

[0049] Based on the above aspects, the embodiments of the present application realize the division of farmland units in farmland regions, construct farmland unit agents, and construct a farmland unit social network for the farmland unit agents, which includes a physical proximity network, a feature similarity network, and a causal correlation network. Through the construction of the physical proximity network, the feature similarity network, and the causal correlation network, the spatial proximity, the feature similarity, and the causal correlation between farmlands are analyzed and quantified from multiple dimensions. The farmland unit social network provides data support for subsequent comprehensive manure regulation strategies.

[0050] The farmland unit agent performs social learning based on the farmland unit social network, obtains individual performance data of the manure regulation strategy, generates a candidate manure regulation strategy set, learns historical efficient strategies through social learning, and generates a candidate manure regulation strategy set in combination with the actual situation of the farmland unit itself, thereby providing a data basis for generating a manure regulation strategy subsequently.

[0051] The candidate manure regulation strategy set is simulated through the causal correlation network and the infectious disease model to obtain strategy radiation data. The influence of each farmland unit in the global farmland region is obtained through the simulation of the strategy radiation, thereby providing a basis for coordinating the local demand and global benefit of the manure regulation strategy subsequently.

[0052] The global farmland unit agent performs global simulation based on the candidate manure regulation strategy set to obtain a strategy global influence index. The global benefit of the farmland after each farmland unit executes the corresponding local optimal strategy without considering the global benefit is quantified through the strategy global influence index, thereby providing a basis for coordinating the local demand and global benefit of the manure regulation strategy subsequently.

[0053] The candidate manure regulation strategy set is optimized according to the individual performance data, the strategy radiation data, and the strategy global influence index to generate an agricultural manure regulation strategy. The individual performance data and the strategy global influence index of the manure regulation strategy are continuously optimized based on the strategy radiation data, so that the finally generated agricultural manure regulation strategy can take into account the local demand and global benefit of the farmland, thereby improving the overall utilization rate of manure and the stability of the farmland ecology. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 is an execution flow schematic diagram of an agricultural manure soil crop interaction optimization regulation method provided by the embodiments of the present application;

[0055] Figure 2 is a schematic diagram of a farmland unit social network in an agricultural manure soil crop interaction optimization regulation method provided by the embodiments of the present application;

[0056] Figure 3is a schematic diagram of an agricultural manure soil crop interaction optimization regulation system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0057] The present application will be described in detail below with reference to the accompanying drawings, Figure 1 is an execution flow schematic diagram of an agricultural manure soil crop interaction optimization regulation method provided by an embodiment of the present application, Figure 2 is a schematic diagram of a farmland unit social network in an agricultural manure soil crop interaction optimization regulation method provided by an embodiment of the present application, which will be described in detail below.

[0058] Step S1, divide the farmland area into farmland units, and construct farmland unit agents.

[0059] Specifically, the continuous farmland area is divided into multiple farmland units of consistent size based on a geographic information system and a clustering algorithm, multiple source data of the multiple farmland units are collected, and the collected raw data is subjected to outlier rejection, missing value filling and data standardization to obtain a farmland unit dataset corresponding to each farmland unit, which at least includes farmland soil data, crop growth data, weather data and farmland management data.

[0060] In a possible embodiment, the farmland area is divided into multiple farmland units of consistent size and farmland characteristics based on a geographic information system and a clustering algorithm, a GIS vector surface segmentation tool is used for farmland unit division, clustering is performed according to farmland soil types, and the farmland area is divided into corresponding farmland unit clusters, such as sandy soil units and clay soil units; clustering is performed according to farmland crop types, and the farmland area is divided into corresponding farmland unit clusters, such as wheat units and corn units; wherein the area of farmland units of the same type is controlled within [0.1, 10] hectares, and unit division is avoided across real barriers such as roads and rivers; specifically, first, the topographic map, soil type distribution map and crop planting planning map corresponding to the farmland area are imported, based on the soil type corresponding boundaries in the soil type distribution map and the crop planting blocks in the crop planting planning map, a clustering algorithm is used to divide the global farmland area into 600 farmland units, including 250 sandy soil wheat units, 150 clay soil wheat units and 200 loam wheat units.

[0061] Collecting farmland soil data through soil sensors, such as collecting the soil nitrogen, phosphorus and potassium contents of each farmland unit through nitrogen, phosphorus and potassium sensors, and collecting soil humidity data in the farmland unit through a soil humidity sensor; using a preset crop growth monitoring camera to monitor the growth period of the farmland crops and obtain corresponding crop growth data; using real-time data from a weather station as weather data; generating corresponding farmland management data according to historical farmland management records, such as the farmland management data in 2021 being {historical fertilization strategy: applying pig manure 15 t / ha, historical yield: wheat yield 6.5 t / ha, farmer decision-making cycle: average 7 days}, and performing standardization processing on the collected data, such as using the KNN algorithm to fill in missing values, using the Z-score method to remove outliers, and mapping the collected data to the [0, 1] interval through min-max standardization to generate a farmland unit dataset;

[0062] Based on the farmland unit dataset and the preset social learning rules and strategy triggering rules, a farmland unit agent is constructed, wherein the social learning rules are used for social learning between the farmland unit and its neighboring farmland units that are physically adjacent or have similar farmland characteristics, such as using a communication interface based on the MQTT protocol to realize information exchange between farmland unit agents; the strategy triggering rules are used for triggering the generation of manure regulation strategies when the farmland unit reaches a predetermined condition, for example, triggering social learning when the soil nitrogen content is less than 80 mg / kg and the farmland crops are in the tillering stage, at which time the farmland unit and its corresponding neighbor farmland units perform social learning to learn the historical optimal manure regulation strategy in the neighbor farmland units, and iteratively optimize the strategy in combination with the farmland unit dataset of the farmland unit itself to generate a corresponding locally optimal manure regulation strategy.

[0063] Step S2, constructing a farmland unit social network for the farmland unit agent, the farmland unit social network including a physical proximity network, a feature similarity network and a causal association network.

[0064] In this embodiment, step S2 includes:

[0065] Step S21, constructing a physical proximity network.

[0066] Specifically, taking the geometric center point of the farmland unit corresponding to the farmland unit agent as a voronoi diagram seed point, constructing a physical proximity network based on the voronoi diagram seed point, and linking the corresponding voronoi diagram seed points according to a pre-set edge connection rule, and using a distance decay function to obtain the edge connection weight.

[0067] In a possible embodiment, the farmland unit agents of the global farmland region are taken as physical proximity network nodes. When the distance between the geometric center points of two farmland units is lower than a preset threshold or there is a common boundary between the plot boundaries of the two farmland units, an undirected edge is established between the two nodes, representing that the two nodes are spatially adjacent, and a distance decay function is used to generate the weight corresponding to the edge connection in the physical proximity network. For example, the distance between the geometric center points of farmland unit A and farmland unit B is 2.8 kilometers, which is lower than the preset threshold of 3 kilometers. Therefore, an undirected edge is established between farmland unit A and farmland unit B, and the edge weight between the two is obtained in the form of an exponential decay function.

[0068] Step S22, constructing a feature similarity network.

[0069] Specifically, the farmland unit agents are taken as nodes of the feature similarity network. Feature extraction is performed on the farmland unit data sets corresponding to the nodes by using a multi-head attention mechanism and a deep residual network structure to generate farmland unit feature vectors, obtain the feature similarity between the farmland unit feature vectors corresponding to the nodes, and establish edge connections in the feature similarity network according to a preset feature similarity threshold. Nodes with edge connections between them are defined as feature similarity nodes.

[0070] In a possible embodiment, feature extraction is performed on the farmland unit data sets of the farmland unit agents to generate corresponding farmland unit feature vectors. The cosine similarity between the farmland unit feature vectors corresponding to the nodes is calculated to quantify the feature similarity of the nodes in the feature similarity network. The five-fold cross-validation method is used to set the feature similarity threshold in the feature similarity network, and whether to establish edge connections between the nodes in the feature similarity network is determined based on the feature similarity threshold. For example, the feature similarity threshold set by the five-fold cross-validation method is 0.75. Assuming that the cosine similarity of the farmland unit feature vectors corresponding to feature similarity network node A and feature similarity network node B is 0.68, the cosine similarity of the farmland unit feature vectors corresponding to feature similarity network node A and feature similarity network node C is 0.78, and the cosine similarity of the farmland unit feature vectors corresponding to feature similarity network node B and feature similarity network node C is 0.85, the edge connection is determined according to the feature similarity threshold of 0.75. It is determined that an edge connection is established between feature similarity network node A and feature similarity network node B, and an edge connection is established between feature similarity network node B and feature similarity network node C.

[0071] Step S23, constructing a causal association network.

[0072] Specifically, the farmland unit agent is taken as a causal correlation network node, a full connection network is established for the causal correlation network node, conditional independence test is performed on the full connection network, edges determined as conditionally independent are removed, the causal correlation strength corresponding to each edge connection in the causal correlation network is synchronously obtained, and the causal correlation network is generated.

[0073] In a possible embodiment, data extraction is performed on the farmland unit data set of the farmland unit agent, and strategy effect correlation data of the last three growth seasons are obtained, the strategy effect correlation data including a fertilization strategy of each farmland unit agent, such as a manure type, an application amount, and an application period, a strategy adoption time, and local effectiveness after implementation of the strategy, such as a yield increase rate and a manure utilization rate, a strategy adoption time of a neighboring farmland unit agent, for example, part of the strategy effect correlation data of the farmland unit A can be represented as {fertilization strategy: [pig manure, 15 t / ha, application in tillering period], strategy adoption time: June 18, xxxx, local effectiveness: [yield increase rate 15%, manure utilization rate 88%], strategy adoption time of neighboring farmland unit: June 28, xxxx}.

[0074] The PC algorithm is used to obtain a causal correlation direction corresponding to a manure regulation strategy between farmland unit agents. Specifically, first, a causal variable set is defined, the causal variable set including a strategy variable X_i of a farmland unit agent i, such as defining the manure application amount as 15 t / ha, X_i=1, and the manure application amount as 10 t / ha, X_i=0, a strategy variable X_j of a farmland unit agent j, a soil nutrient variable S, a meteorological data variable M, and a farmland crop growth period variable G. Fisher transformation is used to determine whether there is a causal correlation between network nodes. For example, it is assumed that the farmland unit A adopts a certain manure scheduling strategy on June 18, xxxx, and the neighboring farmland unit B adopts the same manure scheduling strategy ten days later, after controlling S, M, and G corresponding to the farmland unit A and the farmland unit B, the significance value of the farmland unit A and the farmland unit B is obtained as 0.02 by the Pearson correlation coefficient significance test method, which is lower than the statistical significance judgment standard 0.05, and since the strategy adoption time of the farmland unit A is earlier than that of the farmland unit B, it is determined that there is a causal correlation between the farmland unit A and the farmland unit B, and the corresponding causal correlation direction is that the farmland unit A points to the farmland unit B.

[0075] The Bayesian network is used to estimate the probability that the farmland unit agent j adopts a strategy when the farmland unit agent i adopts a strategy after controlling S, M, and G, and the probability is used as the edge connection weight value between the farmland unit agents i and j, that is, w_{i,j}=P(X_j=1|X_i=1,S,M,G). For example, the edge connection weight between the farmland unit A and the farmland unit B is obtained through the Bayesian network as w_{A,B}=P(X_B=1|X_A=1,S,M,G)=0.78.

[0076] In step S3, the farmland unit agent performs social learning based on the farmland unit social network to obtain individual performance data of the manure regulation strategy and generate a candidate manure regulation strategy set.

[0077] In this embodiment, step S3 includes:

[0078] In step S31, a neighbor farmland unit agent set is constructed.

[0079] Specifically, the farmland unit agent to be regulated by manure is defined as a target farmland unit agent, the corresponding spatial neighbor nodes and feature similar nodes of the target farmland unit agent in the farmland unit social network are obtained, a neighbor similar unit set is constructed, and the corresponding association data set is screened in combination with the causal association network, and the neighbor farmland unit agent set is constructed based on the neighbor similar unit set and the association data set.

[0080] In one possible embodiment, the spatial neighbor nodes that have an edge connection with the target farmland unit agent and have an edge weight not lower than a preset spatial similarity threshold are selected from the physical neighbor network, the feature similar nodes that have an edge connection with the target farmland unit agent and have an edge weight not lower than a preset feature similarity threshold are selected from the feature similar network, and the neighbor similar unit set is constructed based on the spatial neighbor nodes and the feature similar nodes; the causal association nodes that have a causal influence on the target farmland unit agent itself and have an edge weight not lower than a preset causal association threshold are selected from the causal association network to generate an association data set; and the neighbor similar unit set and the association data set are fused and deduplicated to generate the neighbor farmland unit agent set.

[0081] For example, the farmland unit agent A selects 5 spatially adjacent nodes and 7 feature-similar nodes from the physically adjacent network and the feature-similar network, respectively, to construct a set of adjacent and similar units, removes the duplicate 3 feature-similar nodes from the set of adjacent and similar units, and finally retains 5 spatially adjacent nodes and 4 feature-similar nodes. Similarly, the associated data set and the set of adjacent and similar units are removed to generate a set of neighbor farmland unit agents. If the number of elements in the set of neighbor farmland unit agents is less than the lower limit of the preset element quantity threshold, the corresponding threshold is lowered by a preset step, and the corresponding nodes are sequentially expanded into the set of neighbor farmland unit agents according to a preset screening order. The preset screening order is represented as feature-similar node priority > causal association network node priority > spatially adjacent node priority, that is, assuming that the values of the preset spatial similarity threshold, the preset feature similarity threshold, and the preset causal association threshold are 0.8, 0.7, and 0.6, respectively, if the nodes need to be supplemented into the set of neighbor farmland unit agents, the corresponding threshold lower limit can be lowered by 5%. If the number of elements in the set of neighbor farmland unit agents still does not reach the lower limit of the preset element quantity threshold after supplementing the nodes, the threshold lower limit continues to be lowered by 5% until the number of elements in the set of neighbor farmland unit agents reaches the lower limit of the preset element quantity threshold. If the number of elements in the set of neighbor farmland unit agents exceeds the upper limit of the preset element quantity threshold, the redundant elements are removed according to the reverse order of the preset screening order, that is, the farmland unit agent corresponding to the spatially adjacent node with the lowest edge weight in the set of neighbor farmland unit agents is preferentially removed. It should be noted that the number of farmland unit agents corresponding to each type of node in the set of neighbor farmland unit agents, including spatially adjacent nodes, causal association network nodes, and feature-similar nodes, cannot be less than 2. Therefore, when the number of farmland unit agents corresponding to spatially adjacent nodes in the set of neighbor farmland unit agents is 2, the farmland unit agents corresponding to causal association network nodes need to be removed. Similarly, when the number of causal association network nodes is 2, the farmland unit agents corresponding to feature-similar nodes need to be removed.

[0082] It can be understood that when the corresponding threshold is not adjusted to the lower limit of the threshold according to the preset step, the threshold cannot be adjusted below the preset safety boundary. For example, the initial values of the preset spatial similarity threshold, the preset feature similarity threshold, and the preset causal association threshold are 0.8, 0.7, and 0.6, respectively, and the safety boundary is set to 70% of the initial threshold, that is, the adjusted preset spatial similarity threshold, the preset feature similarity threshold, and the preset causal association threshold cannot be lower than 0.56, 0.49, and 0.42, respectively. If the number of elements in the set of neighbor farmland unit agents still does not reach the lower limit of the preset element quantity threshold after adjusting the threshold to the preset safety boundary, the farmland unit agents are screened from the set of neighbor farmland unit agents of the target farmland unit agent and supplemented into the set of neighbor farmland unit agents of the target farmland unit agent.

[0083] For example, the neighbor farmland unit agent set of the farmland unit agent A contains only 5 farmland unit agents, the preset element quantity threshold range is [8, 12], and the number of farmland unit agents contained in the neighbor farmland unit agent set is lower than the lower limit of the preset element quantity threshold, so farmland unit agents need to be supplemented into the neighbor farmland unit agent set. Assuming that the adjustment threshold reaches the preset safety boundary, the number of farmland unit agents contained in the neighbor farmland unit agent set of the farmland unit agent A is 7, at this time, the farmland unit agent B is selected from the neighbor farmland unit agent set, and the farmland unit agents are supplemented from the neighbor farmland unit agent set of the farmland unit agent B according to the preset screening order until the number of elements in the neighbor farmland unit agent set of the target farmland unit agent reaches the lower limit of the preset element quantity threshold.

[0084] In step S32, social learning is performed to generate a candidate manure scheduling strategy set.

[0085] Specifically, the target farmland unit agent learns the historical optimal manure scheduling strategy of the neighbor farmland unit agent to obtain an initial candidate manure scheduling strategy set, the target farmland unit agent asks the neighbor farmland unit agent counterfactually based on the initial candidate manure scheduling strategy set, the neighbor farmland unit agent calls the local causal association network to generate a counterfactual estimation result and sends it to the target farmland unit agent, the counterfactual estimation result is evaluated based on the adversarial review mechanism to obtain an initial candidate manure scheduling strategy evaluation result, and a candidate manure scheduling strategy set is generated.

[0086] It can be understood that after obtaining the initial candidate manure scheduling strategy set, the target farmland unit agent interrogates the neighbor farmland unit agent providing the initial candidate manure scheduling strategy counterfactually. Specifically, the neighbor farmland unit agent calls the corresponding farmland unit data set, synchronously combines the digital twin to construct a virtual scene without implementing the initial candidate manure scheduling strategy, and deduces the growth trajectory and yield result of the farmland crop in the virtual scene to generate a counterfactual estimation result. The counterfactual estimation result is compared with the historical farmland management data in the farmland unit data set to obtain the causal gain value between the strategy execution scene and the non-execution strategy scene. The causal gain value is output as individual performance data, which represents the expected performance of the candidate manure scheduling strategy. For example, when farmland unit agent A recommends the initial candidate manure scheduling strategy of "deep application of 5 tons of manure before the rainy season" to farmland unit agent B, farmland unit agent B will ask farmland unit agent A to show its corresponding historical farmland management record. Assuming that farmland unit agent A implemented the strategy under the same environmental conditions in June last year, the corn yield reached 800 kg / mu, and the counterfactual simulation calculated that in the scenario without using the strategy, the corn yield was only 650 kg / mu, and the corresponding causal gain value was 150 kg / mu.

[0087] In one possible embodiment, the historical performance data corresponding to all farmland unit agents in the farmland unit social network is obtained, the historical performance data is represented as the deviation value of the actual performance and the expected performance of the historical manure scheduling strategy, and the farmland unit agents are ranked in ascending order of the historical performance data. A predetermined number of farmland unit agents are selected from the highest and lowest ranked historical performance data, respectively, to generate a candidate counter-opinion agent set. For example, the top 5 farmland unit agents with the highest historical performance data and the bottom 3 farmland unit agents with the lowest historical performance data are selected to form the candidate counter-opinion agent set. The intermediate centrality of the farmland unit agents in the candidate counter-opinion agent set in the farmland unit social network is obtained, and the farmland unit agents within a predetermined threshold range are selected to generate a counter-opinion agent set, where the intermediate centrality is represented as the number of times a node is located on the shortest path between other nodes. The more times, the more important the node is in the network. For example, the predetermined threshold range of the intermediate centrality is [0.7, 0.85]. Assuming that only the intermediate centrality of farmland unit agent B in the candidate counter-opinion agent set is 0.6, which is lower than the lower limit of the predetermined threshold range of the intermediate centrality, the candidate counter-opinion agent is removed, and the remaining candidate counter-opinion agents form the counter-opinion agent. The counter-fact estimation result is input into the counter-opinion agent set for prediction evaluation to generate an initial candidate manure scheduling strategy evaluation result. Based on the initial candidate manure scheduling strategy evaluation result and combined with the propensity score algorithm, a candidate manure scheduling strategy is selected from the initial candidate manure scheduling strategy set to generate a candidate manure scheduling strategy set. For example, when generating the candidate manure scheduling strategy set of farmland unit A, the "whether to adopt a certain strategy" is used as the dependent variable, and the farmland unit data set of farmland unit A is used as the independent variable. The propensity scores of the 6 initial candidate manure scheduling strategies obtained through social learning are calculated by Logistic regression. Assuming that the propensity scores of the 6 initial candidate manure scheduling strategies are 0.72, 0.71, 0.68, 0.64, 0.61, and 0.58, respectively, the strategies with propensity scores lower than 0.65 are removed, and the remaining initial candidate manure scheduling strategies are packaged as a candidate manure scheduling strategy set for output.

[0088] In step S4, the candidate manure regulation strategy set is simulated by the causal correlation network and the infectious disease model to obtain strategy radiation data.

[0089] Specifically, the diffusion process of the candidate manure regulation strategy set in the global farmland unit corresponding ecosystem is simulated based on the causal correlation network and the infectious disease model to obtain the strategy radiation data of each candidate manure regulation strategy in the candidate manure regulation strategy set. The strategy radiation data is represented as the radiation propagation influence degree of the candidate manure regulation strategy in the global farmland area after the implementation of the candidate manure regulation strategy in a certain farmland unit.

[0090] It can be understood that based on the infectious disease model, the nodes in the causal correlation network are divided into an infection period, a latent period and an immune period, the infection period is represented as the adoption and implementation of the candidate manure regulation strategy by the farmland unit, the latent period is represented as the adoption but non-implementation of the candidate manure regulation strategy by the farmland unit, and the immune period is represented as the implementation of the candidate manure regulation strategy by the farmland unit without expected effectiveness, and the farmland unit no longer accepts the candidate manure regulation strategy within a preset immune period.

[0091] In a possible embodiment, a plurality of nodes in the causal correlation network are randomly selected as infected nodes in the infection period in the infectious disease model, a transmission rate is set as a conversion probability for the transmission infection of uninfected nodes, so that the uninfected nodes enter the latent period, the nodes entering the latent period are defined as latent nodes, an illness rate is set as a conversion probability for the uninfected nodes to be converted into infected nodes in the infection period, and an immunity rate is set as a conversion probability for the infected nodes to be converted into immune nodes in the immune period; the strategy radiation simulation is performed based on the Monte Carlo algorithm, and the strategy radiation degree data of the farmland unit corresponding to the nodes in the causal correlation network are generated.

[0092] It can be understood that the transmission rate is represented as the probability that the farmland unit j also adopts the candidate manure regulation strategy after the farmland unit i adopts the candidate manure regulation strategy, wherein the transmission rate is in a positive correlation with the edge weight of the causal correlation network, for example, the transmission rate = 1.2 * w_{i,j}, wherein 1.2 is a preset transmission rate coefficient, and the preset transmission rate coefficient is used to reduce the deviation between the transmission rate in the strategy radiation simulation and the actual transmission rate of the strategy, that is, assuming that the edge connection weight between the farmland unit i and the farmland unit j is 0.5, the transmission rates corresponding to the two are 0.6; the illness rate is represented as the probability that the farmland unit adopts the candidate manure regulation strategy from knowing the candidate manure regulation strategy, and the illness rate is generated based on the historical farmer decision period included in the farmland unit data set of the farmland unit, for example, assuming that the historical average decision period of a farmland unit is 5 days, the illness rate = 1 / 5 = 0.20 / day, if the candidate manure regulation strategy involves complex operations such as mixed manure ratio, the illness rate is reduced to 75% of the initial value, that is, the illness rate corresponding to the candidate manure regulation strategy is adjusted to 0.15 / day; the immunity rate is represented as the probability that the farmland unit rejects the same strategy after adopting the candidate manure regulation strategy, and is set based on the probability that the farmland unit in the farmland unit data set does not adopt the same strategy in the future 2 growth seasons when the implementation effect of the strategy does not reach the expectation, for example, after a farmland unit adopts manure regulation strategy X, the actual implementation effect does not reach the expectation, the maximum similarity of the manure regulation strategies adopted by the farmland unit in the subsequent 2 growth seasons to manure regulation strategy X is 60%, and the immunity rate of the farmland unit is set to 0.6.

[0093] Specifically, the global candidate manure regulation strategy set corresponding to each farmland unit is screened according to a preset target candidate strategy screening rule, and a target candidate strategy for strategy radiation simulation is obtained. For example, according to the target candidate strategy screening rule that "a candidate manure regulation strategy is determined as a target candidate strategy if it is included in the candidate manure regulation strategy set of at least 20% of the global farmland units and the average treatment effect corresponding to the strategy is not less than 9%", the candidate manure regulation strategy set of farmland unit A is screened, and the following three target candidate strategies are screened out: target candidate strategy S1 {pig manure single application}, target candidate strategy S2 {pig manure, chicken manure and cow manure are applied in proportion}, and target candidate strategy S3 {pig manure, chicken manure and cow manure are applied in rotation}.

[0094] The state space of the strategy radiation simulation is initialized, and simulation is performed for each target candidate strategy. First, the original unit corresponding to each target candidate strategy is selected, which represents the farmland unit that generates the strategy earliest and has the highest average treatment effect. Then, the direct influence unit of the original unit in the causal correlation network is obtained, which represents the farmland unit that has an edge connection with the node corresponding to the original unit in the causal correlation network and has an edge weight not less than a preset threshold. For example, the farmland unit corresponding to the causal correlation node with an edge connection weight not less than 0.6 with the node corresponding to the original unit can be regarded as a direct influence unit. A predetermined number of global farmland units are randomly selected, and the state of the selected farmland units is initialized to the infection period. The farmland units that are not initialized to enter the infection period are set to the susceptible state, which represents that the farmland unit has not received any candidate manure regulation strategy.

[0095] Further, multi-time step iterative simulation is adopted. In each time step, the farmland unit in the infection period infects its neighbor farmland unit with the transmission rate corresponding to the farmland unit and its neighbor farmland unit as the conversion probability, so that the neighbor farmland unit enters the latent period. The farmland unit in the latent period enters the infection period with the disease rate as the conversion probability. For the farmland unit in the infection period, the strategy execution state of the current farmland unit is randomly determined to be "strategy failure, not as expected" by a predetermined rule. For the farmland unit determined to be strategy failure, the state of the farmland unit is converted to the immune period with the immune rate corresponding to the farmland unit as the conversion probability. For example, the strategy execution state of the farmland unit is randomly determined by taking the historical failure rate average corresponding to the currently executed candidate manure regulation strategy as the determination rule. The above simulation process is repeated until each farmland unit completes at least one simulation process, and the basic reproduction number, the final propagation size and the average propagation delay of each farmland unit after simulation are obtained.

[0096] generate a global radiation influence heat map based on the strategy radiation index, and output the radiation influence heat map as strategy radiation data. Specifically, the basic reproductive number, the final size of the propagation, and the average propagation delay are weighted and summed according to a predetermined weight distribution ratio to generate the strategy radiation index corresponding to each farmland unit. For example, assuming that the basic reproductive number, the final size of the propagation, and the average propagation delay obtained by farmland unit A after simulation are 2.0, 0.8, and 0.6 respectively, and the strategy radiation index obtained after weighted summation according to a ratio of 4:3:3 is 1.22.

[0097] It can be understood that the basic reproductive number can be represented as At this time, the basic reproductive number means that when farmland unit i is taken as the initial infectious period node, the average number of newly infected farmland units that can be activated by a single infectious period farmland unit in the simulation process, and when , it indicates that the candidate manure regulation strategy corresponding to unit i has sustainable propagation, The greater the value, the higher the propagation radiation intensity of the strategy and the greater the influence of the farmland unit in the global farmland area. For example, when the of farmland unit A is 2.5, it means that when each farmland unit in the infectious period adopts the candidate manure regulation strategy provided by farmland unit A, an average of 2.5 farmland units can be infected to enter the infectious period; the final size of the propagation can be represented as At this time, the final size of the propagation means that after the simulation ends, the total number of farmland units in the strategy propagation chain of farmland unit i that enter the infectious period accounts for the proportion of the global farmland units. The greater the value of the final size of the propagation, the greater the radiation range of the candidate manure regulation strategy corresponding to the farmland unit; the average propagation delay can be represented as At this time, the average propagation delay means the average time from activation to propagation to all farmland units in the infectious period of the candidate manure regulation strategy corresponding to farmland unit i, and the average propagation delay is standardized. For example, the average propagation delay of the candidate manure regulation strategy of farmland unit E is 10 days, and the maximum delay time of the global is 20 days, and the standardized value = 1-10 / 20 = 0.5.

[0098] In step S5, the global farmland unit agent performs global simulation based on the set of candidate manure regulation strategies to obtain a global influence index of the strategy.

[0099] Specifically, the global farmland unit agent matches a corresponding candidate manure regulation strategy from the candidate manure regulation strategy set, and obtains a strategy global influence index based on a strategy global influence indicator, the strategy global influence indicator at least including an absorption efficiency indicator, an environmental impact indicator, and an economic benefit indicator.

[0100] In a possible embodiment, each farmland unit agent selects a locally optimal strategy from the candidate manure regulation strategy set of itself as a simulation execution strategy, the locally optimal strategy being represented as a strategy with the highest average treatment effect, each farmland unit implements the locally optimal strategy in a pre-constructed simulation environment to simulate a complete natural growing season, and obtains an absorption efficiency indicator, an environmental impact indicator, and an economic benefit indicator of each farmland unit when implementing the corresponding locally optimal strategy, and generates a strategy global influence index according to the absorption efficiency indicator, the environmental impact indicator, and the economic benefit indicator.

[0101] It can be understood that the absorption efficiency indicator is used to quantify the absorption efficiency of soil nutrients in the global farmland unit after the global farmland unit agent executes the corresponding candidate manure regulation strategy; the environmental impact indicator is used to quantify the influence rate of the global farmland on the environment after the global farmland unit agent executes the corresponding candidate manure regulation strategy; and the economic benefit indicator is used to quantify the fluctuation of the expected yield of crops in the global farmland after the global farmland unit agent executes the corresponding candidate manure regulation strategy.

[0102] For example, the global farmland unit includes three farmland units, farmland unit A, farmland unit B and farmland unit C, and the corresponding local optimal strategies are strategy A, strategy B and strategy C respectively. Through a complete natural growth season simulation, assuming that the total nitrogen application amount of manure is 50t and the total nitrogen absorption amount of crops is 45t, the corresponding absorption efficiency is 90%, and the nitrogen absorption efficiency is output as the absorption efficiency index; assuming that the soil types of farmland unit A, farmland unit B and farmland unit C are only sandy soil and clay soil, the average water loss amount of the sandy soil unit and the clay soil unit is compared with the preset natural loss standard, and the water loss rates of the sandy soil unit and the clay soil unit are obtained as 80% and 60% respectively, the water loss rate is weighted and summed to obtain the final water loss rate of 70%, and the water loss rate is output as the environmental impact index; assuming that the average yield of the three farmland units A, B and C is increased by 20% compared with the historical yield after adopting the corresponding local optimal strategy, the value of the economic benefit index is 0.2; the values of the absorption efficiency index, the environmental impact index and the economic benefit index obtained by the preset weight ratio are weighted and summed to obtain the strategy global influence index of farmland unit A, farmland unit B and farmland unit C, which is 0.9*0.3+0.7*0.4+0.2*0.3=0.61 (here, only the acquisition of the strategy global influence index is exemplified, and the specific determination needs to be combined with the actual situation).

[0103] Step S6, optimizing the candidate manure regulation strategy set according to the individual performance data, the strategy radiation data and the strategy global influence index, and generating the agricultural manure regulation strategy.

[0104] Specifically, based on the strategy radiation data, the individual performance data of each candidate manure regulation strategy in the candidate manure regulation strategy set and the strategy global influence index of the candidate manure regulation strategy set are coordinated to generate the agricultural manure regulation strategy.

[0105] In a possible embodiment, the global farmland is divided into a high radiation degree region, a medium radiation degree region and a low radiation degree region based on the strategy radiation degree data, the weight proportion of the divided regions is adaptively adjusted, and the corresponding strategy comprehensive score is calculated according to the adjusted weight proportion. For example, the farmland units in the region where the strategy radiation degree index in the radiation influence thermal map is not less than 0.8 are divided into the high radiation degree region, the farmland units in the region where the strategy radiation degree index is in [0.6, 0.8] are divided into the medium radiation degree region, and the farmland units in the region where the strategy radiation degree index is less than 0.6 are divided into the low radiation degree region. The corresponding weight is set by using the analytic hierarchy process combined with the characteristics of the divided regions. For example, in the high radiation degree region, the driving effect of strategy propagation on the global is focused, and the weight proportion of the individual performance data, the strategy radiation degree data and the strategy global influence index is 1:2:1. In the medium radiation degree region, the balance of the three indexes of the individual performance data, the strategy radiation degree data and the strategy global influence index is focused, and the weight proportion of the individual performance data, the strategy radiation degree data and the strategy global influence index is 1:1:1. In the low radiation degree region, the combination of the local benefits of the farmland units and the global benefits of the farmland is emphasized, and the weight proportion of the individual performance data, the strategy radiation degree data and the strategy global influence index is 2:1:2.

[0106] The three candidate manure regulation strategies with the highest strategy comprehensive scores in the candidate manure regulation strategy set of each farmland unit are obtained. If the number of candidate manure regulation strategies is less than three, all the candidate manure regulation strategies are directly called. The non-dominated sorting genetic algorithm is used, and the sum of the strategy comprehensive scores of all farmland units is used as the optimization objective function to obtain the global Pareto optimal solution set of the farmland. The normalized values of the absorption efficiency index, the environmental impact index and the economic benefit index corresponding to each strategy in the global Pareto optimal solution set of the farmland are obtained, and the normalized values are defined as the membership degrees. The normalized method is the actual value of the index / the maximum value of the index. The closer the value of the membership degree is to 1, the better the strategy performs in the corresponding index dimension. The comprehensive membership degrees of the corresponding regions are obtained combined with the weight proportions of the high radiation degree region, the medium radiation degree region and the low radiation degree region. The solution with the highest comprehensive membership degree in the corresponding region is selected as the global comprehensive optimal strategy, and the global comprehensive optimal strategy is output as the agricultural manure regulation strategy.

[0107] Figure 3 is a schematic diagram of an agricultural manure soil crop interaction optimization regulation system provided by an embodiment of the present application.

[0108] Specifically, an agricultural manure soil crop interaction optimization regulation system comprises:

[0109] a unit division module, configured to divide a farmland region into farmland units and construct farmland unit agents.

[0110] a network construction module, configured to construct a farmland unit social network including a physical proximity network, a feature similarity network and a causal correlation network for the farmland unit agents.

[0111] a social learning module, configured to perform social learning by the farmland unit agents, acquire individual performance data of manure regulation strategies, and generate a candidate set of manure regulation strategies.

[0112] a radiation simulation module, configured to perform strategy radiation simulation on the candidate set of manure regulation strategies according to the causal correlation network and an infectious disease model, and generate strategy radiation degree data.

[0113] a global simulation module, configured to perform global simulation on the candidate set of manure regulation strategies by global farmland unit agents, and acquire a strategy global influence index.

[0114] a strategy generation module, configured to optimize the candidate set of manure regulation strategies according to the individual performance data, the strategy radiation degree data and the strategy global influence index, and generate an agricultural manure regulation strategy.

[0115] The specific use and role of the embodiment are described as follows:

[0116] First, the farmland region is divided into farmland units to construct farmland unit agents, and a farmland unit social network including a physical proximity network, a feature similarity network and a causal correlation network is constructed for the farmland unit agents. By constructing the physical proximity network, the feature similarity network and the causal correlation network, the spatial proximity, the feature similarity and the causal correlation among farmlands are analyzed and quantified from multiple dimensions, and the farmland unit social network provides a data basis for subsequent comprehensive manure regulation strategies.

[0117] Then, the farmland unit agents perform social learning based on the farmland unit social network to acquire individual performance data of manure regulation strategies and generate a candidate set of manure regulation strategies. The candidate set of manure regulation strategies is generated by learning historical efficient strategies through social learning and combining the actual situation of the farmland unit, and provides data support for measuring the local demand and benefit of the farmland unit for subsequent generation of manure regulation strategies.

[0118] Then, the candidate manure regulation strategy set is simulated by the causal network and the infectious disease model to obtain strategy radiation data, and the influence of each farmland unit in the global farmland area is obtained by the simulation, which provides a connection bridge for coordinating the local demand and global benefit of the manure regulation strategy.

[0119] Then, the global farmland unit agent performs global simulation based on the candidate manure regulation strategy set to obtain a global influence index of the strategy, which quantifies the global benefit of the farmland after each farmland unit executes the corresponding local optimal strategy without considering the global benefit, thereby providing a basis for measuring the global effect of the farmland for coordinating the local demand and global benefit of the manure regulation strategy.

[0120] Finally, the candidate manure regulation strategy set is optimized according to the individual performance data, strategy radiation data, and strategy global influence index to generate an agricultural manure regulation strategy, and the individual performance data and strategy global influence index of the manure regulation strategy are continuously optimized based on the strategy radiation data, so that the finally generated agricultural manure regulation strategy can balance the local demand and global benefit of the farmland, thereby improving the overall utilization rate of manure and the stability of the farmland ecology.

[0121] It should be understood that the term "and / or" herein merely describes the association relationship of the associated objects, and can represent three relationships, for example, A and / or B, which can represent three cases of A alone, A and B together, and B alone, where A and B can be singular or plural. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects, but can also represent an "and / or" relationship, which can be understood according to the context before and after.

[0122] It should be understood that in the embodiments of the present application, the size of the serial number of the above processes does not mean the order of execution, and the execution order of the processes should be determined according to their functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0123] The above-described embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for optimizing and regulating the interaction between agricultural manure, soil, and crops, characterized in that, The method includes: S1: Divide the farmland area into farmland units and construct farmland unit intelligent agents; Based on geographic information systems and clustering algorithms, continuous farmland areas are divided into multiple farmland units of uniform size. Multi-source data are collected from multiple farmland units, and outlier removal, missing value imputation, and data standardization are performed on the collected raw data to obtain a farmland unit dataset corresponding to each farmland unit. The farmland unit dataset includes at least farmland soil data, crop growth data, meteorological data, and farmland management data. Based on the farmland unit dataset and the preset social learning rules and policy triggering rules, a farmland unit intelligent agent is constructed. The social learning rules are used for farmland units to learn socially with neighboring farmland units that are physically adjacent or have similar farmland characteristics. The policy triggering rules are used for farmland units to trigger the generation of manure regulation strategies when they meet predetermined conditions. S2: Construct a farm unit social network for farm unit agents. The farm unit social network includes a physical proximity network, a feature similarity network, and a causal association network. S3: The farmland unit agent performs social learning based on the farmland unit social network to obtain individual efficacy data of manure regulation strategies and generate a candidate manure regulation strategy set. The farmland unit agent to be regulated by manure is defined as the target farmland unit agent. The spatial neighbor nodes and feature similar nodes corresponding to the target farmland unit agent in the farmland unit social network are obtained to construct a set of neighboring similar units. The corresponding associated dataset is then filtered by combining the causal association network. Based on the set of neighboring similar units and the associated dataset, a set of neighboring farmland unit agents is constructed. The target farmland unit agent learns the historical best manure scheduling strategy from the neighboring farmland unit agents to obtain an initial candidate manure scheduling strategy set. The target farmland unit agent performs counterfactual questioning on the neighboring farmland unit agents based on the initial candidate manure scheduling strategy set. The neighboring farmland unit agents call the local causal association network to generate counterfactual estimation results and send them to the target farmland unit agent. Obtain historical performance data for all farmland unit agents in the farmland unit social network. The historical performance data represents the deviation between the actual performance and the expected performance of the historical manure scheduling strategy. Select a predetermined number of farmland unit agents from the highest and lowest ranked historical performance data respectively to generate a candidate adversarial evaluation agent set. Obtain the middle centrality of farmland unit agents within the candidate adversarial evaluation agent set in the farmland unit social network, select farmland unit agents within a predetermined threshold range, and generate an adversarial evaluation agent set; The counterfactual estimation results are input into the set of adversarial evaluation agents for prediction and evaluation, generating initial candidate manure scheduling strategy evaluation results. Based on the initial candidate manure scheduling strategy evaluation results, a set of candidate manure scheduling strategies and corresponding individual performance data are generated, and the individual performance data represents the expected performance of the candidate manure scheduling strategy. S4: Use causal relationship networks and infectious disease models to simulate the radiation of candidate manure regulation strategies and obtain strategy radiation data; S5: The intelligent agent of the whole-domain farmland unit performs whole-domain simulation based on the candidate manure regulation strategy set to obtain the whole-domain impact index of the strategy. S6: Optimize the candidate manure regulation strategy set based on individual efficacy data, strategy radiation data, and strategy global impact index to generate agricultural manure regulation strategies.

2. The method for optimizing and regulating the interaction between agricultural manure, soil, and crops according to claim 1, characterized in that, To construct a farmland unit social network for farmland unit agents, the farmland unit social network includes physical proximity networks, feature similarity networks, and causal association networks, including: Using the geometric center point of the farmland unit corresponding to the farmland unit agent as the seed point of the Veno diagram, a physical proximity network is constructed based on the seed point of the Veno diagram, and the corresponding seed points of the Veno diagram are linked according to the pre-set edge connection rules. The corresponding edge connection weights are obtained by using the distance decay function. Using farmland unit agents as feature similarity network nodes, feature extraction is performed on the farmland unit dataset corresponding to each node through multi-head attention mechanism and deep residual network structure to generate farmland unit feature vectors, obtain the feature similarity between the farmland unit feature vectors corresponding to each node, and establish edge connections of the feature similarity network according to the preset feature similarity threshold. Nodes with edge connections between nodes are defined as feature similar nodes. Using farmland unit agents as nodes in a causal network, a fully connected network is established for these nodes. Conditional independence is tested on the fully connected network, edges that are determined to be conditionally independent are removed, and the causal correlation strengths corresponding to the connections of each edge in the causal network are obtained simultaneously to generate the causal network.

3. The method for optimizing and regulating the interaction between agricultural manure and soil crops according to claim 1, characterized in that, The candidate manure regulation strategy set was simulated using causal association networks and infectious disease models to obtain strategy radiation data, including: The diffusion process of the candidate manure regulation strategy set in the corresponding ecosystem of the whole farmland unit was simulated based on the causal relationship network and infectious disease model, and the strategy radiation data of each candidate manure regulation strategy in the candidate manure regulation strategy set were obtained. The strategy radiation data represents the degree of radiation and spread of the candidate manure regulation strategy within a certain farmland unit to the entire farmland area.

4. The method for optimizing and regulating the interaction between agricultural manure, soil, and crops according to claim 3, characterized in that, The method further includes: Based on the infectious disease model, nodes in the causal association network are divided into the infection period, the incubation period, and the immunity period; The infection period refers to the farmland unit adopting and implementing the candidate manure control strategy; the incubation period refers to the farmland unit adopting but not implementing the candidate manure control strategy; and the immunity period refers to the farmland unit implementing the candidate manure control strategy but failing to achieve the expected effectiveness, and not accepting the candidate manure control strategy again during the preset immunity period. Multiple causal network nodes are randomly selected as infected nodes in the infectious disease model during the infection period. The transmission rate is set as the conversion probability to spread infection to uninfected nodes, causing uninfected nodes to enter the incubation period. At the same time, nodes that enter the incubation period are defined as latent nodes. Uninfected nodes are set as the conversion probability to enter the infection period during the infection period during the lesion rate. Infected nodes are set as the conversion probability to enter the immune period during the immunity rate. Policy radiation simulation was performed using the Monte Carlo algorithm to generate policy radiation data for farmland units corresponding to causal network nodes.

5. The method for optimizing and regulating the interaction between agricultural manure, soil, and crops according to claim 1, characterized in that, The intelligent agent of the entire farmland unit performs a global simulation based on a set of candidate manure regulation strategies to obtain the global impact index of the strategies, including: The intelligent agent of the whole-domain farmland unit matches the corresponding candidate manure regulation strategy from the candidate manure regulation strategy set, and obtains the whole-domain impact index of the strategy through simulation. The whole-domain impact index of the strategy includes at least absorption efficiency index, environmental impact index and economic benefit index. The whole-domain impact index of the strategy is obtained based on the whole-domain impact index of the strategy. The absorption efficiency index is used to quantify the absorption efficiency of soil nutrients in the entire farmland after the intelligent agent of the whole farmland unit executes the corresponding candidate manure regulation strategy. Environmental impact indicators are used to quantify the environmental impact rate of the entire farmland after the intelligent agent of the whole farmland unit executes the corresponding candidate manure control strategy; The economic return indicator is used to quantify the fluctuation of the expected return of crops in the entire farmland area after the intelligent agent of the whole farmland unit executes the corresponding candidate manure control strategy.

6. The method for optimizing and regulating the interaction between agricultural manure and soil crops according to claim 1, characterized in that, Based on individual efficacy data, strategy radiation data, and the strategy global impact index, the candidate manure regulation strategy set is optimized to generate agricultural manure regulation strategies, including: Based on the strategy radiation data, the individual efficacy data of each candidate manure regulation strategy in the candidate manure regulation strategy set and the strategy global influence index of the candidate manure regulation strategy set are coordinated to generate agricultural manure regulation strategies.

7. An agricultural manure-soil-crop interaction optimization and regulation system, used to implement the method described in any one of claims 1 to 6, characterized in that, include: A unit division module is used to divide farmland areas into farmland units and construct farmland unit intelligent agents. A network construction module is used to construct a farmland unit social network for farmland unit agents, including a physical proximity network, a feature similarity network, and a causal association network. The social learning module is used by farmland unit agents to perform social learning, acquire individual efficacy data of manure regulation strategies, and generate a candidate manure regulation strategy set. The radiation simulation module is used to perform strategy radiation simulation on a set of candidate manure regulation strategies based on causal relationship networks and infectious disease models, and generate strategy radiation data. The global simulation module is used by the global farmland unit agent to perform global simulation based on the candidate manure regulation strategy set, and obtain the global impact index of the strategy. The strategy generation module is used to optimize the candidate manure regulation strategy set based on individual effectiveness data, strategy radiation data, and strategy global influence index, and generate agricultural manure regulation strategies.

Images