Cultivation management and agricultural machinery management and control method and system based on large-model multi-Agent and Agent RAG

Abstract

The invention discloses a large-model multi-Agent and Agent RAG-based cultivation management and agricultural machinery management and control method and system, and the method comprises the steps: obtaining crop habitat basic data covering the whole stage of cultivation management and harvesting through a sensor or inspection perception; by analyzing planting specifications and data, operation standards are set; a task planning total Agent and a task triggering Agent are constructed, a man-machine interaction visual interface is designed, and task planning is autonomously generated according to farmland conditions; constructing data Agents, repairing abnormal data, and making a scheduling decision through Agentive RAG under the condition that basic data is lost or extreme; constructing a knowledge-based mechanism decision model, and generating an agricultural implement decision and management method through the acquired data; according to the decision information of the agricultural implement, the operation specification of the agricultural implement is adjusted, the operation condition is presented through a visual interface, and the planting data and the decision information are stored in a database; according to the invention, crop cultivation knowledge processing capability in a complex agricultural scene is realized, and agricultural machinery autonomous scheduling of each operation link of agricultural production is realized.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of agricultural machinery decision-making management and control, and particularly relates to a cultivation management and agricultural machinery management and control method and system based on a large model multi-agent and Agentic RAG, which can realize autonomous decision-making of agricultural machinery operation scheduling according to crop habitat data and cultivation management knowledge. BACKGROUND

[0002] In the process of modern agricultural mechanization production, farmers rely on traditional experience to schedule agricultural machinery operation, which not only leads to low efficiency of agricultural machinery resource use and increased operating costs, but also may cause economic losses due to decision-making errors or failure to schedule in time. Even in the current unmanned farm operation and maintenance, due to the lack of autonomous cultivation management and agricultural machinery scheduling system, timely agricultural machinery operation scheduling still relies on manual issuance of operation scheduling tasks.

[0003] In modern agricultural production, sensors are widely used in farmland monitoring to collect key information about soil state, environmental state and crop state. These data constitute the data basis for supporting crop cultivation management and agricultural machinery operation decision-making. However, due to the large amount and complex structure of these habitat data, manual analysis is often time-consuming and labor-intensive. In addition, when sensors fail or unexpected events occur in farmland, these habitat data cannot effectively support the decision-making of agricultural machinery operation scheduling in a timely manner.

[0004] In order to improve the automation and intelligent management level of agricultural production and enhance the autonomy of cultivation management and agricultural machinery scheduling in unmanned farms, it is urgent to invent an autonomous cultivation management and agricultural machinery management and control method and system to process perception data and cultivation knowledge in agricultural production, optimize agricultural machinery resource allocation strategies according to changes in field data, automatically generate crop planting, field management and harvesting scheduling decisions, and improve the robustness and intelligent management level of autonomous decision-making systems for cultivation management and agricultural machinery scheduling. SUMMARY

[0005] The present application provides a cultivation management and agricultural machinery management and control method and system based on a large model multi-agent and Agentic RAG, which can realize autonomous decision-making of agricultural machinery operation scheduling according to crop habitat data and cultivation management.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme:

[0007] The structure of the cultivation management and agricultural machinery management and control system based on a large model multi-agent and Agentic RAG of the present application comprises the following substructures:

[0008] (a) External data part, including crop habitat basic data and agricultural machinery information and material information, crop planting specifications and historical planting information;

[0009] (b) Cultivation management and agricultural machinery control decision model part, including task planning total Agent, task triggering Agent, data Agent, decision information database, network and planting knowledge database and mechanism decision model based on knowledge graph structure and associated knowledge;

[0010] (c) Generation information part, including output decision information and natural language conversion machine instruction;

[0011] The crop habitat basic data is divided into farmland environment information, soil information and crop information, wherein the farmland environment information includes air temperature and humidity, total light duration, weather in five days, rainfall and wind speed and the like; the soil information includes soil temperature and humidity, hardness, nitrogen, phosphorus and potassium content and the like; the crop information includes crop phenotype and growth state and the like; the agricultural machinery information includes agricultural machinery function, type and quantity and the like; the material information includes seed, pesticide, fertilizer and the like agricultural input product inventory information;

[0012] The crop planting specification and historical planting information are each planting standard of specific crops, wherein the crop planting specification includes specific information such as seeding amount, base fertilizer amount and planting spacing, and the historical planting information includes planting standards and yield of previous years and the like.

[0013] The application is based on the cultivation management and agricultural machinery control method of large model multi-Agent and Agentic RAG, and the cultivation management and agricultural machinery control system based on the foregoing large model multi-Agent and Agentic RAG is applied, including the following steps:

[0014] S1: through arranging field sensor Internet of Things or inspection sensing mode, obtaining crop habitat basic data of crops, soil and field microclimate in each stage of crop growth, the data covering four stages of crop cultivation, management and harvesting, and these basic data will be used for agricultural machinery operation scheduling decision basis;

[0015] S2: through large model analysis of crop planting specification and historical planting data, obtaining application material and operation specification, including seed amount, base fertilizer amount and agricultural machinery operation range index and the like information, and setting operation standard according to the obtained information;

[0016] S3: constructing task planning total Agent and task triggering Agent, designing man-machine interaction visual interface based on natural language dialogue system, analyzing according to the current farm situation, generating task instruction, and generating detailed task planning according to the instruction;

[0017] S4: constructing data Agent, filtering and repairing abnormal data value, and in addition, making scheduling decision data through Agentic RAG in the case of loss or extreme situation of crop habitat basic data;

[0018] S5: Construct a mechanism decision model based on the knowledge graph structure and associated knowledge. Based on the obtained crop habitat basic data or data generated by data Agents, generate a task list for the plot, and generate operation agricultural machinery decision and management method recommendations combined with agricultural machinery and material information;

[0019] S6: Adjust the agricultural machinery operation parameters according to the agricultural machinery decision information, including carrying machinery, materials, oil quantity, and operation range indicators, and adjust the sowing quantity and pesticide and fertilizer application quantity according to the management method and operation standard to present the agricultural machinery operation and material use of each plot in natural language, and then convert the natural language into machine instructions that can be recognized by agricultural machinery and agricultural robots, and complete the operation tasks of each plot according to the machine instructions, and then save the planting data and decision information to the database.

[0020] As an optimal solution for cultivation management and agricultural machinery control method based on large model multi-Agent and Agentic RAG, the task planning total Agent understands and analyzes the data holding situation and generates scheduling instructions. This method uses a variant of the supervisor architecture, and the task planning total Agent is the supervisor in the multi-Agent system, responsible for the unified management of this method, using the task triggering Agent and data Agents as tools, and also including a decision model;

[0021] As an optimal solution for cultivation management and agricultural machinery control method based on large model multi-Agent and Agentic RAG, the task triggering Agent is used to determine the start and continuation nodes of the cultivation management and harvesting tasks, and to release the required items in the agricultural machinery and material warehouse in advance, and to record the details of the daily operation. The process includes:

[0022] (a) The task triggering Agent uses the date tool to determine the nodes of the crop planting stage, and combines with the weather forecast tool to determine the future workable dates, and records the key date nodes through the data recording tool;

[0023] (b) After determining the operation date, the retrieval query tool is used to query the network or historical database to obtain the required materials for this operation task, and transmit them to the agricultural machinery and material warehouse for preparation;

[0024] (c) On the day when the operation date is reached, the task triggering Agent sends a task execution signal to the task planning total Agent, and the task planning total Agent plans the tasks for the day and uses data Agents and decision models according to the plan;

[0025] (d) After the daily operation is completed, record the operation details, including task completion information, task incomplete information, missing materials and agricultural machinery information, etc.

[0026] As the preferred solution of the cultivation management and agricultural machinery management and control method based on large model multi-agent and Agentic RAG, the data Agents include environmental data Agents, data query Agents, prediction decision Agents and data fitting Agents, wherein the data query Agents, the prediction decision Agents and the data fitting Agents, together with the historical decision database and the network database, constitute the Agentic RAG;

[0027] The environmental data Agents are used to correct or complete the data when the sensor outputs error data or missing data, and the process includes:

[0028] (a) data screening is performed by an outlier detection tool to determine the part to which the abnormal data belongs;

[0029] (b) there are abnormalities in sowing, harvesting and use of pesticides and fertilizers, and the abnormalities are corrected by a state correction tool;

[0030] (c) there are environmental data abnormalities, and the environmental information is obtained by a weather forecast tool to correct or complete the abnormalities;

[0031] (d) there are soil data abnormalities, and the soil information is obtained by an inference model tool to correct or complete the abnormalities;

[0032] The data query Agents are used to query network or historical data to generate network data information, and the process includes:

[0033] (a) crop information analysis and growth stage calculation are performed by a crop growth stage analysis tool;

[0034] (b) the original network data information is generated by querying the information under similar crop stages and environmental conditions in the network or historical data by a search query tool;

[0035] (c) the network data information is obtained by data cleaning and preprocessing of the original network data information by a data processing tool;

[0036] The prediction decision Agents are used to integrate multi-aspect information to make the next step decision, and the process includes:

[0037] (a) the environmental information and soil information at the latest time are extracted by a database search tool;

[0038] (b) the data information is integrated with the network data information by a data integration tool;

[0039] (c) the decision is made according to the integrated data by a decision tool;

[0040] Wherein the data fitting Agent is used to generate fitting data or prediction data using prompt words when encountering emergencies, and is substituted into the decision model to play a preventive or remedial function, improving the robustness of the decision-making process, the process includes:

[0041] (a) generating prompt words that meet the current environment according to decision information through a prompt word generation tool

[0042] (b) inputting the prompt words into the LLM through a text processing tool and processing the resulting LLM output text;

[0043] (c) processing the resulting text data through a data generation tool to obtain data in a format suitable for the decision model;

[0044] As an optimal solution for the cultivation management and agricultural machinery management and control method based on large model multi-Agent and Agentic RAG, the mechanism decision model based on knowledge graph structure and associated knowledge gives the operation of agricultural machinery and the recommended management method according to the input data, wherein the input data covers the complete cultivation management and harvesting operation stage, including environmental information and soil information in the cultivation stage, and crop information is added in the management and harvesting stage, and the decision model will also be dynamically adjusted according to different operation stages, including starting a multi-modal crop disease and pest model in the management and harvesting stage to process crop information, the process includes:

[0045] (a) generating a task list that needs to be performed on the current plot according to the obtained data, and setting priorities, when there is missing information or prediction is needed, these information can be generated by environmental data Agent or prediction decision Agent;

[0046] (b) generating a recommended list of agricultural machinery according to the plot task list, and setting priorities according to the function composition, fuel consumption, etc. of the agricultural machinery;

[0047] (c) selecting according to the obtained agricultural machinery and material information combined with the agricultural machinery list;

[0048] (d) when the agricultural machinery is successfully assigned, the carrying equipment and materials, as well as the operation range, application amount, etc. are given, and the operation time is calculated according to the plot area and operation speed; when the agricultural machinery is not successfully assigned, the plot enters the waiting assignment stage, and when other agricultural machinery is completed, it is re-assigned;

[0049] As an optimal solution for the cultivation management and agricultural machinery management and control method based on large model multi-Agent and Agentic RAG, the final agricultural machinery decision result is interacted in a natural language manner, and an instruction is generated to complete the scheduling of agricultural machinery, the process includes:

[0050] (a) presenting the agricultural machine decision information in a natural language manner, including carrying machine, material, oil amount, operation range index, and operation specification of pesticide and fertilizer application;

[0051] (b) converting the natural language into machine instructions recognizable by the agricultural machine and the agricultural robot, and completing the agricultural machine allocation task of each plot according to the machine instructions;

[0052] (c) storing the planting data and decision information after the task is completed in the database;

[0053] The beneficial effects of the present application are:

[0054] (1) The present application adopts multiple large model Agents to process crop habitat basic data, including but not limited to soil and environmental information data, crop state and pest monitoring data. Using the autonomous planning and thinking chain processing capability of the large model multi-Agent, the above-mentioned farmland and agricultural machine data can be analyzed, and customized planting schemes, fine management strategies and optimal harvesting time prediction can be dynamically generated. Compared with the traditional analysis mode relying on artificial experience, the present application reduces the investment of human resources, alleviates the problems of artificial decision lag and one-sidedness caused by information processing delay and knowledge barriers, and realizes scientific agricultural management.

[0055] (2) The present application uses Agentic RAG to realize highly dynamic task management and multi-step reasoning in complex agricultural scenarios. Using the adaptive retrieval and continuous learning capability of Agentic RAG, uncertain events such as crop habitat basic data anomalies or sudden conditions of farmland are actively processed, data defects are automatically identified, retrieval strategies are adjusted in time, the problem of decreased robustness of the decision system caused by data source failure or external environment mutation is alleviated, and the knowledge processing capability in complex agricultural scenarios is improved.

[0056] (3) The present application can generate appropriate agricultural machine resource scheduling strategies according to the operation task demand, operation plot distribution, agricultural material reserve information, and agricultural machine state by using the dynamic planning capability of the large model multi-Agent and the knowledge retrieval and reasoning capability of Agentic RAG at the agricultural machine decision level. The strategy guarantees the operation efficiency and reduces the operation cost. In addition, in the complex collaborative task of multiple plots, the present application can realize unified scheduling and efficient collaboration of multiple agricultural machines, improve the equipment utilization rate, and reduce the driving and waiting time. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only exemplary, and for those skilled in the art, other drawings can be obtained from the provided drawings without creative labor;

[0058] The structures, proportions, sizes, etc. shown in the specification are only used to cooperate with the content disclosed in the specification, to be understood and read by those skilled in the art, and are not used to limit the limiting conditions that the present application can be implemented, so they do not have technical significance. Any modification of structure, change of proportion relationship or adjustment of size, without affecting the effect and purpose that the present application can produce, should still fall within the scope of the technical content disclosed by the present application;

[0059] Figure 1 The cultivation management and agricultural machinery control system structure schematic diagram based on large model multi-agent and Agentic RAG provided by the present application is shown in the following figure:

[0060] Figure 2 The cultivation management and agricultural machinery control method flowchart provided by the present application based on large model multi-agent and Agentic RAG is shown in the following figure:

[0061] Figure 3 The detailed decision process schematic diagram of the cultivation management and agricultural machinery control method provided by the present application based on large model multi-agent and Agentic RAG is shown in the following figure:

[0062] Figure 4 The mechanism decision model flowchart of the cultivation management and agricultural machinery control method provided by the present application based on large model multi-agent and Agentic RAG based on knowledge graph structure and associated knowledge is shown in the following figure:

[0063] Figure 5 The cultivation management and agricultural machinery control method generation flowchart provided by the present application in embodiment 1 is shown in the following figure:

[0064] Figure 6 The cultivation management and agricultural machinery control method generation flowchart provided by the present application in embodiment 2 is shown in the following figure:

[0065] Figure 7 The cultivation management and agricultural machinery control system flowchart provided by the present application based on large model multi-agent and Agentic RAG is shown in the following figure:

[0066] Figure 8The multi-agent autonomous planning interface in the agricultural task intelligent planning system designed by the cultivation management and agricultural machinery management and control method based on large model multi-agent and Agentic RAG provided by the application;

[0067] Figure 9 The agricultural machinery resource deployment interface in the agricultural task intelligent planning system designed by the cultivation management and agricultural machinery management and control method based on large model multi-agent and Agentic RAG provided by the application. DETAILED DESCRIPTION

[0068] The embodiments of the present application are described below by specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0069] Referring to Figure 1 The application provides the composition structure of the cultivation management and agricultural machinery management and control system based on large model multi-agent and Agentic RAG, including the following substructures:

[0070] (a) External data part, including crop habitat basic data and agricultural machinery information and material information, crop planting specifications and historical planting information;

[0071] (b) Cultivation management and agricultural machinery management and control decision model part, including task planning general agent, task triggering agent, data agent, decision information database, network and planting knowledge database and mechanism decision model based on knowledge graph structure and associated knowledge;

[0072] (c) Generated information part, including output decision information and natural language converted machine instructions;

[0073] Referring to Figure 2 The application provides a cultivation management and agricultural machinery management and control method based on large model multi-agent and Agentic RAG, including the following steps:

[0074] S1: Obtain crop habitat basic data of crops, soil and field microclimate at each stage of crop growth by arranging field sensor Internet of Things or inspection sensing, and the data covers the four stages of crop cultivation, management and harvesting. These basic data will be used as the basis for agricultural machinery operation scheduling decision;

[0075] S2: Obtain the application material and operation specification, including seed dosage, base fertilizer dosage, and agricultural machinery operation range index information, by analyzing crop planting specification and historical planting information through a large model, and set operation standards according to the obtained information;

[0076] S3: Construct a task planning total Agent and a task triggering Agent, design a man-machine interactive visual interface based on a natural language dialogue system, analyze the current farm situation, generate task instructions, and generate detailed task planning according to the instructions;

[0077] S4: Construct data Agents, filter and repair abnormal data values, and make scheduling decision data in the case of loss or extreme situation of crop habitat basic data through Agentic RAG;

[0078] S5: Construct a mechanism decision model based on a knowledge graph structure and associated knowledge, generate a task list for the plot based on the obtained crop habitat basic data or data generated by the data Agents, and generate operation agricultural implement decision and management method recommendation combined with agricultural implement and material information;

[0079] S6: According to the agricultural implement decision information, adjust the agricultural implement operation parameters, including carrying implements, materials, oil quantity, operation range index, and adjust the operation specification such as seeding rate and pesticide and fertilizer application rate according to the management method and operation standard, present the agricultural implement operation and material use situation of each plot in a natural language manner, then convert the natural language into machine instructions recognizable by agricultural implements and agricultural robots, and complete the operation tasks of each plot according to the machine instructions, and then save the planting data and decision information to the database.

[0080] In step S1, crop habitat basic data of crops, soil, and field microclimate at each stage of crop growth is obtained through arrangement of field sensor Internet of Things or inspection sensing, and the data covers the four stages of crop cultivation, management, and harvesting. These basic data will be used as the basis for agricultural implement operation scheduling decision;

[0081] Specifically, the crop habitat basic data includes farmland environment information, soil information, and crop information. The farmland environment information includes air temperature and humidity, total illumination time, weather in the past five days, rainfall, and wind speed, etc. The soil information includes soil temperature and humidity, hardness, nitrogen, phosphorus, and potassium content, etc. The crop information includes crop phenotype and growth state, etc. The material information includes seed, pesticide, fertilizer, and other agricultural input product inventory information;

[0082] In step S2, the application material and operation specification, including seed dosage, base fertilizer dosage, and agricultural machinery operation range index information, are obtained by analyzing crop planting specification and historical planting information through a large model, and operation standards are set according to the obtained information;

[0083] Specifically, the crop planting specification and the historical planting information are various planting standards for specific crops, wherein the crop planting specification includes specific information such as seeding amount, base fertilizer amount, planting spacing, and the historical planting information includes planting standards and yield of previous years;

[0084] In step S3, the task planning total Agent and the task triggering Agent are constructed, a man-machine interactive visual interface is designed based on a natural language dialogue system, the current farm situation is analyzed, a task instruction is generated, a detailed task plan is generated according to the instruction, and the following steps are included:

[0085] (a) The task planning total Agent issues a task instruction, the task triggering Agent uses a date tool to determine various nodes of the crop planting stage, and determines future workable dates in combination with a weather forecast tool, and records key date nodes through a data recording tool;

[0086] (b) After determining the work dates, the network or the historical database is queried through a retrieval query tool to obtain the materials required for this work task, and is transmitted to the agricultural machinery and material warehouse for preparation;

[0087] (c) On the day when the work date is reached, the task triggering Agent sends a task execution signal to the task planning total Agent, and the task planning total Agent plans the task of the day, and uses data Agents and decision models according to the plan;

[0088] (d) After the daily work is completed, the work details are recorded, including task completion information, task incomplete information, missing materials and agricultural machinery information, etc.;

[0089] In step S4, the data Agents are constructed, and abnormal data values are filtered and repaired, and in addition, the Agentic RAG makes scheduling decision data in the case of loss of basic data of crop habitat or extreme situation

[0090] Specifically, as shown in Figure 3 The data Agents include an environmental data Agent, a data query Agent, a prediction decision Agent and a data fitting Agent, wherein the data query Agent, the prediction decision Agent and the data fitting Agent, together with a historical decision database and a network database, constitute the Agentic RAG;

[0091] The environmental data Agent is used to correct or complete data when the sensor outputs error data or missing data, and includes the following steps:

[0092] (a) The data is screened through an abnormal value detection tool to determine the part to which the abnormal data belongs;

[0093] (b) There are abnormalities in sowing, harvesting and fertilizer status, which are corrected by state correction tools;

[0094] (c) There are environmental data abnormalities, which are corrected or supplemented by weather forecast tools to obtain environmental information;

[0095] (d) There are soil data abnormalities, which are corrected or supplemented by inference model tools to obtain soil information;

[0096] Among them, the data query Agent is used to query network or historical data, and generate network data information, including the following steps:

[0097] (a) Through the crop growth stage analysis tool to analyze crop information and calculate the growth stage;

[0098] (b) Query information under similar crop stage and environmental conditions in network or historical data through the search query tool to generate original network data information;

[0099] (c) Through the data processing tool, the original network data information is cleaned and preprocessed to obtain network data information;

[0100] Among them, the prediction decision Agent is used to integrate multi-aspect information and make the next step decision, including the following steps:

[0101] (a) Through the database search tool, the environmental information and soil information of the recent time are searched and the data information is extracted;

[0102] (b) Through the data integration tool, the data information and network data information are integrated;

[0103] (c) Through the decision tool, the decision is made according to the integrated data;

[0104] Among them, the data fitting Agent is used to generate fitting data or prediction data using prompt words when encountering unexpected events, and is used to prevent or remedy the function, improve the robustness of the decision-making process, including the following steps:

[0105] (a) Through the prompt word generation tool, the prompt word conforming to the current environment is generated according to the decision information;

[0106] (b) Through the text processing tool, the prompt word is input into LLM and the LLM output text is processed;

[0107] (c) Through the data generation tool, the text data is processed and the data format suitable for the decision model is obtained;

[0108] In step S5, a mechanism decision model based on the knowledge graph structure and associated knowledge is constructed, a task list is generated for the plot based on the obtained crop habitat basic data or data generated by the data Agents, and an operation agricultural implement decision and management method recommendation is generated in combination with the agricultural implement and material information;

[0109] Specifically, as shown in Figure 4 the following steps are included:

[0110] (a) According to the obtained data, a task list currently required to be executed by the plot is generated, and a priority is set. When there is missing information or prediction is required, the information can be generated by the environmental data Agent or the prediction decision Agent;

[0111] (b) According to the plot task list, a list of recommended agricultural implements is generated, and a priority is set according to the function composition, fuel consumption, etc. of the agricultural implements;

[0112] (c) According to the obtained agricultural implement and material information, selection is made in combination with the agricultural implement list;

[0113] (d) When the agricultural implement is successfully assigned, the carrying equipment and materials, operation range, application and fertilization amount, etc. are given, and the operation time is calculated according to the plot area and operation speed; when the agricultural implement fails to be assigned, the plot enters the waiting assignment stage, and is re-assigned when there is another completed operation agricultural implement;

[0114] In step S6, according to the agricultural implement decision information, the operation parameters of the agricultural implement are adjusted, including the carrying implement, material, oil amount, operation range index, etc., and the operation specifications such as seeding amount and application and fertilization amount are adjusted according to the management method and operation standard. In a natural language manner, the agricultural implement operation and material use of each plot are presented, then the natural language is converted into machine instructions recognizable by the agricultural implement and agricultural robot, and the operation tasks of each plot are completed according to the machine instructions, then the planting data and decision information are saved to the database, including the following steps:

[0115] (a) The agricultural implement decision information is presented in a natural language manner, including suggestions on the carrying implement, material, oil amount, operation range index, application and fertilization amount, etc. operation specifications;

[0116] (b) The natural language is converted into machine instructions recognizable by the agricultural implement and agricultural robot, and the agricultural implement assignment tasks of each plot are completed according to the machine instructions;

[0117] (c) The planting data and decision information after the task completion are saved to the database.

[0118] Example 1

[0119] In embodiment 1, specific cultivation management and agricultural machinery control method generation examples are provided as follows:

[0120] After arranging the field sensor Internet of Things or inspection sensing, obtaining the crop habitat basic data and operation standards of white radish, the operation of the white radish cultivation and sowing stage is started;

[0121] As shown in Figure 5 , first, the task planning total Agent plans the task, the PlannerAgent splits the subtasks, including triggering related tasks, calculating operation dates, environmental data detection, agricultural machinery decision and recording decision details; the subtasks are handed over to the ReActAgent to independently select appropriate tools for execution, each subtask returns a result and the task is re-evaluated by the PlannerAgent, wherein the numerical number is the step sequence, including the following steps:

[0122] T1: The task triggering Agent obtains the instruction issued by the task planning total Agent: start preparing the operation of land preparation and sowing; and the PlannerAgent splits the subtasks, including date query, weather information query, material information query and data recording; the subtasks are handed over to the ReActAgent to independently select appropriate tools for execution, each subtask returns a result and the task is re-evaluated by the PlannerAgent;

[0123] T2: The date tool queries the current date and determines the time node of land preparation and sowing, combines the weather forecast tool to determine the future workable date, and records the key date node through the data recording tool; after determining the operation date, the network or historical database is queried through the retrieval query tool to obtain the required materials for this operation task, and the required materials and agricultural machinery are transmitted to the material warehouse and agricultural machinery prompt;

[0124] T3: On the day of the operation date, the task triggering Agent sends an operation signal to the task planning total Agent, and the task planning total Agent plans the task for the day;

[0125] T4: The environmental data Agent obtains the instruction issued by the task planning total Agent: operation; and the PlannerAgent splits the subtasks, including anomaly detection, anomaly state correction, weather information query and inference fitting data; the subtasks are handed over to the ReActAgent to independently select appropriate tools for execution, each subtask returns a result and the task is re-evaluated by the PlannerAgent;

[0126] T5: The outlier detection tool performs data screening, and then the outlier detection result is returned to the PlannerAgent to re-evaluate the task. If there is no outlier, the subsequent task is cancelled. If there is an outlier, the task is updated according to the category of the outlier. The correct state information, environmental information or fitted soil information of the outlier or missing value is obtained through the state correction tool, weather forecasting tool or reasoning model tool, and the data is corrected or completed;

[0127] T6: After repairing or completing all outliers, the environmental data Agent sends a data normal signal to the task planning total Agent, and the task planning total Agent plans the remaining tasks;

[0128] T7: The task planning total Agent calls the decision model of the related task to make a decision. Based on the repaired data, the task list is generated for the plot, and the operation agricultural implement decision and management method recommendation are generated in combination with the agricultural implement and material information;

[0129] T8: The decision information is summarized and the operation details are recorded, including task completion information, task incomplete information, missing material and agricultural implement information, etc., and then saved into the decision information database.

[0130] Example 2

[0131] In Example 2, a specific cultivation management and agricultural machine control method generation example is provided as follows:

[0132] After arranging the field sensor Internet of Things or inspection sensing, obtaining the crop habitat basic data and operation standards of white radish, the preventive management operation of the field management stage of white radish is started;

[0133] As shown in Figure 6 , first, the task planning total Agent plans the task, the PlannerAgent splits the subtasks, including triggering related tasks, calculating operation dates, data query, generating decision information, generating simulation data and recording decision details. The subtasks are executed by the ReActAgent independently selecting appropriate tools. Each subtask returns a result and the task is re-evaluated by the PlannerAgent, wherein the numerical number is the step sequence, including the following steps:

[0134] T1: The task triggering Agent obtains the instruction issued by the task planning total Agent: to perform the field management task; and the PlannerAgent splits the subtasks, including date query, weather warning query and data recording. The subtasks are executed by the ReActAgent independently selecting appropriate tools. Each subtask returns a result and the task is re-evaluated by the PlannerAgent;

[0135] T2: The weather forecast tool queries whether there is any meteorological warning information recently, and the data recording tool displays the current crop planting situation, planting days, etc.

[0136] T3: The task planning total agent makes preventive management measures according to the planting days and meteorological warnings, and the data query agent obtains the instructions issued by the task planning total agent: data query, calculates the current growth stage of the crop through the crop growth stage analysis tool, then queries the network or historical data for information under similar crop stages and environmental conditions through the search query tool, generates original network data information, and then processes the original network data information through the data processing tool to obtain network data information;

[0137] T4: The task planning total agent makes preventive management measures according to the obtained network data information, and the prediction decision agent obtains the instructions issued by the task planning total agent: generate decision information, retrieve the latest environmental information and soil information through the database retrieval tool, extract data information, then integrate the data information and network data information through the data integration tool, and finally make decisions based on the integrated data through the decision tool;

[0138] T5: The task planning total agent makes preventive management measures according to the decision information, and the data fitting agent obtains the instructions issued by the task planning total agent: generate fitting data, generates prompt words that meet the current environment through the prompt word generation tool according to the decision information, then inputs the prompt words into the LLM and processes the obtained LLM output text through the text processing tool, and finally processes the obtained text data through the data generation tool and obtains data in a format suitable for the decision model;

[0139] T6: The task planning total agent calls the decision model of the relevant task to make decisions, generates a task list for the plot based on the fitted basic data, and generates operation agricultural machinery decision and management method recommendations combined with agricultural machinery and material information

[0140] T7: The decision information is summarized and recorded for operation details, including task completion information, task incomplete information, missing material and agricultural machinery information, etc., and then saved into the decision information database;

[0141] The above Agent is an intelligent agent supported by a large model, and has the abilities of task planning, task decomposition, adaptation to external specifications, reflection, optimization and memory. In summary, the application obtains crop habitat basic data of crops, soil and field microclimate in each stage of crop growth through arranging field sensor Internet of Things or inspection sensing, and the data covers the four stages of crop cultivation, management and harvesting. These basic data will be used as the basis for agricultural machinery operation scheduling decision; through large model analysis of crop planting specifications and historical planting information, the application obtains application materials and operation specifications, including seed dosage, base fertilizer dosage and agricultural machinery operation range index information, and sets operation standards according to the obtained information; a task planning total Agent and a task triggering Agent are constructed, a man-machine interactive visual interface is designed based on a natural language dialogue system, task instructions are generated according to the current farm situation, and detailed task planning is generated according to the instructions; data Agents are constructed, abnormal data values are filtered and repaired, and scheduling decision data is made by Agentic RAG in the case of loss of crop habitat basic data or extreme conditions; a mechanism decision model based on a knowledge graph structure and associated knowledge is constructed, a task list is generated for a plot according to the obtained crop habitat basic data or the data generated by the data Agents, and operation agricultural machinery decision and management method recommendation is generated in combination with agricultural machinery and material information; according to the agricultural machinery decision information, the operation parameters of the agricultural machinery are adjusted, including the carrying machinery, material, oil quantity, operation range index and the like, and the operation specifications such as seeding amount and pesticide and fertilizer application amount are adjusted according to the management method and operation standard, and the operation of the agricultural machinery and the use of the material of each plot are presented in a natural language manner. Then, the natural language is converted into machine instructions recognizable by agricultural machinery and agricultural robots, and the operation tasks of each plot are completed according to the machine instructions, and then the planting data and decision information are saved to the database. The application adopts multiple Agents to coordinate and cooperate to ensure effective execution of tasks, each Agent is responsible for processing of a specific task, can dynamically adjust the scheme according to the needs of the planting stage, and realizes the whole process planning goal through cooperation. The application has intelligence and self-learning, and the Agent model can learn through feedback and optimize itself. The multiple Agents and Agentic RAG can realize automatic adjustment of data and autonomous allocation of operation tasks according to real-time farmland data state, network and planting knowledge, historical decision records and task requirements, so as to improve the robustness of the intelligent decision system, and finally achieve dynamic optimization of agricultural production and intelligent management of agricultural machinery.

[0142] It should be noted that the method of the embodiment can be executed by a single device, such as a computer or a server. The method of the embodiment can also be applied to a distributed scenario, and completed by multiple devices cooperating with each other. In the case of a distributed scenario, one of the multiple devices can only execute one or more steps in the method of the embodiment, and the multiple devices will interact with each other to complete the method.

[0143] It is to be understood that the embodiments which have been described are merely possible examples and other embodiments can be within the scope of the appended claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve desirable results. Additionally, the processes depicted in the attached figures do not necessarily require the particular order shown, or sequential order to achieve desirable results. In certain implementations, multitasking and parallel processing can be advantageous.

[0144] Referring to Figure 6 The present application provides a cultivation management and agricultural machinery control system based on large model multi-agent and Agentic RAG, comprising:

[0145] A data acquisition and processing module M1 is configured to acquire crop habitat basic data of crops, soil and field microclimate at each stage of crop growth through field sensor Internet of Things or inspection sensing, and the data covers the four stages of crop cultivation, management and harvesting. These basic data will be used as the basis for agricultural machinery operation scheduling decision-making;

[0146] An operation standard setting module M2 is configured to obtain application materials and operation standards, including seed dosage, base fertilizer dosage and agricultural machinery operation range index information, by analyzing crop planting specifications and historical planting information through a large model, and set operation standards according to the obtained information;

[0147] A task planning total agent and task triggering agent construction and processing module M3 is configured to design a man-machine interactive visual interface based on a natural language dialogue system, analyze the current farm situation, generate task instructions, and generate detailed task planning according to the instructions;

[0148] A data agent construction and processing module M4 is configured to filter and repair abnormal data values, and make scheduling decision data in the case of loss of crop habitat basic data or extreme conditions through Agentic RAG;

[0149] A decision model construction module M5 is configured to construct a mechanism decision model based on a knowledge graph structure and associated knowledge, generate a task list for a plot based on the acquired crop habitat basic data or data generated by the data agent, and generate agricultural machinery operation decision and management method recommendations in combination with agricultural machinery and material information;

[0150] A natural language conversion module M6 is configured to present the tasks and agricultural machinery operation situation of each plot, including suggestions on operation specifications such as carrying machinery, materials, oil quantity, operation range index, and application amount of pesticides and fertilizers, and then convert the natural language into machine instructions that can be recognized by agricultural machinery and agricultural robots, and complete the operation tasks of each plot according to the machine instructions.

[0151] In this system, the basic crop habitat data in the data acquisition and processing module M1 is divided into farmland environmental information, soil information, and crop information. Farmland environmental information includes air temperature and humidity, total sunshine duration, weather data for the past five days, rainfall, and wind speed; soil information includes soil temperature and humidity, hardness, and nitrogen, phosphorus, and potassium content; and crop information includes crop phenotype and growth status.

[0152] In this system, the crop planting specifications and historical planting information in the operation standard setting module M2 are the various planting standards for the specific crop. The crop planting specifications include specific information such as seeding rate, base fertilizer application rate, and planting spacing, while the historical planting information includes planting standards and yields from previous years.

[0153] In this system, the task planning master agent and task triggering agent construction and processing module M3 are used. The task planning master agent understands and analyzes the data holding status and generates scheduling instructions. This module adopts a variant of the supervisor architecture. The task planning master agent acts as the supervisor in the multi-agent system, responsible for the unified management of this method. It uses task triggering agents and data agents as tools, and also includes a decision model.

[0154] The task triggering agent is used to determine the start and continuation nodes of the planting, management, and harvesting tasks, and to release agricultural machinery and materials needed for the warehouse in advance. In addition, it records the details of the day's work. The process includes:

[0155] (a) The task planning agent issues task instructions, the task triggering agent uses date tools to determine each node of the crop planting stage, and combines weather forecast tools to determine the future workable dates, and records the key date nodes through data recording tools;

[0156] (b) After determining the operation date, search the network or historical database using search tools to find out the materials required for this operation and transmit them to the agricultural machinery and material warehouse for preparation;

[0157] (c) On the day the task is due, the task triggering agent sends a task execution signal to the task planning agent, and the task planning agent plans the tasks for the day and uses data agents and decision models according to the plan.

[0158] (d) After the daily work is completed, record the details of the work, including information on task completion, information on tasks not completed, and information on missing materials and agricultural machinery, etc.

[0159] In the system, the Agents in the data Agents construction and processing module M4 include environment data Agents, information query Agents, prediction decision Agents, and data fitting Agents, wherein the information query Agents, the prediction decision Agents, and the data fitting Agents, together with the historical decision database and the network database, constitute the Agentic RAG;

[0160] The environment data Agents are used to correct or complete data when the sensors output error data or missing data, and the process includes:

[0161] (a) data screening is performed through an outlier detection tool to determine the part to which the abnormal data belongs;

[0162] (b) there are abnormalities in seeding, harvesting, and the use of fertilizers and pesticides, and the abnormalities are corrected through a state correction tool;

[0163] (c) there are environmental data abnormalities, and environmental information is obtained through a weather forecast tool to correct or complete the abnormalities;

[0164] (d) there are soil data abnormalities, and soil information is obtained through an inference model tool to correct or complete the abnormalities;

[0165] The information query Agents are used to query network or historical data to generate network information, and the process includes:

[0166] (a) crop information analysis and growth stage calculation are performed through a crop growth stage analysis tool;

[0167] (b) original network information is generated by querying information under similar crop stages and environmental conditions in network or historical data through a search query tool;

[0168] (c) network information is obtained by performing data cleaning and preprocessing on the original network information through a data processing tool;

[0169] The prediction decision Agents are used to integrate multiple aspects of information to make the next decision, and the process includes:

[0170] (a) environmental information and soil information in the recent time are extracted through a database search tool;

[0171] (b) data information is integrated with network information through a data integration tool;

[0172] (c) a decision is made according to the integrated data through a decision tool;

[0173] Wherein the data fitting Agent is used to generate fitting data or prediction data using the prompt words when encountering an emergency, and substitute into the decision model, playing a preventive or remedial function, improving the robustness of the decision-making process, the process includes:

[0174] (a) generating prompt words in accordance with the decision information using the prompt word generation tool;

[0175] (b) inputting the prompt words into the LLM and processing the obtained LLM output text through the text processing tool;

[0176] (c) processing the obtained text data and obtaining a data format suitable for the decision model through the data generation tool;

[0177] In the system, the mechanism decision model based on knowledge graph structure and associated knowledge constructed by the decision model construction module M5 gives recommendations for job agricultural implement decision and management methods according to input data, wherein the input data covers complete cultivation management and harvesting job stages, such as environment information and soil information in the cultivation stage, and crop information is added in the management and harvesting stage, and the decision model will also be dynamically adjusted according to different job stages, such as starting a multi-modal crop disease and pest model in the management and harvesting stage to process crop information, the process includes:

[0178] (a) generating a task list that needs to be performed on the current plot according to the obtained data, and setting priorities, when there is missing information or prediction is needed, these information can be generated by the environment data Agent or the prediction decision Agent;

[0179] (b) generating a list of recommended agricultural implements according to the plot task list, and setting priorities according to the function composition, fuel consumption, etc. of the agricultural implements;

[0180] (c) selecting according to the obtained agricultural implement and material information in combination with the agricultural implement list;

[0181] (d) when the agricultural implement is successfully assigned, the carrying equipment and materials, as well as the recommended job speed, pesticide amount, etc. are given, and the plot job time is generated according to the plot area and job speed; when the agricultural implement fails to be assigned, the plot enters the waiting assignment stage, and when other job completed agricultural implements are available, the plot is re-assigned;

[0182] In the system, the natural language conversion module M6 enables the agricultural implement decision result to be interacted in a natural language manner, and generates instructions to complete the scheduling of the agricultural implement, the process includes:

[0183] (a) presenting the agricultural implement decision information in a natural language manner, including carrying implements, materials, oil amount, job range indicators, pesticide and fertilizer application amount, etc. suggestions on job specifications;

[0184] (b) converting the natural language into machine instructions recognizable by the agricultural machine and the agricultural robot, and completing the agricultural machine allocation task of each plot according to the machine instructions;

[0185] (c) storing the planting data and decision information after the task is completed into the database;

[0186] It should be noted that the information interaction and execution process between the modules of the system described above are based on the same concept as the method embodiment in Embodiment 1 of the present application, and the technical effects brought about are the same as those of the method embodiment.

[0187] The present application provides a non-transitory computer-readable storage medium, which stores program codes of a cultivation management and agricultural machine management and control method based on large model multi-Agent and Agentic RAG, the program codes comprising instructions for executing the cultivation management and agricultural machine management and control method based on large model multi-Agent and Agentic RAG of Embodiment 1 or any possible implementation manner thereof.

[0188] The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media sets. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk, SSD), etc.

[0189] The present application provides an electronic device, comprising: a memory and a processor;

[0190] The processor and the memory complete mutual communication through a bus; the memory stores program instructions executable by the processor, and the processor calling the program instructions can execute the cultivation management and agricultural machine management and control method based on large model multi-Agent and Agentic RAG of Embodiment 1 or any possible implementation manner thereof.

[0191] Specifically, the processor can be implemented by hardware or software, when implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor, which reads software codes stored in the memory to implement, the memory can be integrated in the processor or located outside the processor and exist independently.

[0192] In the above embodiments, all or part can be realized by software, hardware, firmware or any combination thereof. When realized by software, all or part can be realized in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed by a computer, all or part generates the processes or functions described in the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another via wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.).

[0193] To verify the effectiveness of the cultivation management and agricultural machinery control method and system based on large model multi-agent and Agentic RAG of the present application, autonomous cultivation management and agricultural machinery scheduling systems for different crops and different planting links are developed and experimental verification is carried out. As shown in Figure 8 As shown in FIG. 6, by analyzing the incoming crop habitat basic data and agricultural machinery and inventory material information, it can be seen that the large model multi-agent and Agentic RAG can realize autonomous planning of each operation link in the planting process. As shown in Figure 9 According to the planning information and agricultural machinery decision information, suitable agricultural resource allocation is generated. The experimental results prove that the present application can process farmland data and crop cultivation knowledge in complex agricultural scenarios, and realize the autonomous scheduling capability of agricultural machinery in each operation link of agricultural production.

[0194] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present application can be realized by a general computing system, which can be concentrated on a single computing system or distributed on a network composed of multiple computing systems, and optionally, they can be realized by program codes executable by a computing system, so that they can be stored in a storage system and executed by a computing system, and in some cases, the steps shown or described herein can be executed in different order, or they can be manufactured into individual integrated circuit modules, or multiple modules or steps can be manufactured into a single integrated circuit module, so that the present application is not limited to any specific hardware and software combination.

[0195] Although the present application has been described in detail with general description and specific embodiments above, it is obvious to those skilled in the art that some modifications or improvements can be made on the basis of the present application. Therefore, these modifications or improvements made on the basis of not deviating from the spirit of the present application, all belong to the scope of protection claimed by the present application.

Claims

1. A cultivation management and agricultural machinery control system based on large-scale multi-agent and agentic RAG models, characterized in that, Includes the following substructures: (a) External data section, including basic crop habitat data, agricultural machinery and material information, crop planting standards and historical planting information; (b) The decision-making model for cultivation management and agricultural machinery control includes the task planning agent, task triggering agent, data agents, decision information database, network and planting knowledge database, and mechanism decision-making model based on knowledge graph structure and related knowledge; (c) The information generation part, including the output decision information and machine instructions converted from natural language; The basic data on crop habitats are divided into farmland environmental information, soil information, and crop information. Farmland environmental information includes air temperature and humidity, total sunshine duration, weather conditions over the past five days, rainfall, and wind speed. Soil information includes soil temperature and humidity, hardness, and nitrogen, phosphorus, and potassium content. Crop information includes crop phenotypic and growth status data. Agricultural machinery information includes the functions, types, and quantities of agricultural machinery. Material information includes seed, pesticide, and fertilizer inventory information. The crop planting specifications and historical planting information refer to the various planting standards for the specific crop. The crop planting specifications include specific information on sowing amount, base fertilizer application amount and planting spacing, while the historical planting information includes planting standards and yields for each year.

2. A cultivation management and agricultural machinery control method based on large-scale multi-agent and agentic RAG models, characterized in that, The cultivation management and agricultural machinery control system based on large-scale multi-agent and agentic RAG as described in claim 1 includes the following steps: S1: By deploying field sensors, IoT, or inspection and sensing methods, basic data on crop habitat, soil, and field microclimate at each stage of crop growth are obtained. The data covers the four stages of crop cultivation, management, and harvesting. This basic data will be used as the basis for agricultural machinery operation scheduling decisions. S2: By analyzing crop planting standards and historical planting information through large-scale model analysis, the application materials and operation standards are obtained, including seed usage, base fertilizer usage and agricultural machinery operation range indicators, and operation standards are set based on the obtained information. S3: Construct a task planning agent and a task triggering agent, design a human-computer interaction visualization interface based on a natural language dialogue system, analyze the current farm situation, generate task instructions, and generate detailed task plans based on the instructions; S4: Build data agents to filter and repair abnormal data values. In addition, make scheduling decisions through Agentic RAG in case of loss of basic crop habitat data or extreme situations. S5: Construct a mechanism decision-making model based on knowledge graph structure and related knowledge. By acquiring basic crop habitat data or data generated by data agents, generate a task list for each plot of land, and combine agricultural machinery and material information to generate recommendations for agricultural machinery operation decisions and management methods. S6: Based on agricultural machinery decision information, adjust agricultural machinery operation parameters, including the amount of implements, materials, fuel, and operating range indicators. Adjust seeding rate, pesticide and fertilizer application rates according to management methods and operation standards. Present the agricultural machinery operation and material usage of each plot in natural language. Then, convert the natural language into machine instructions that can be recognized by agricultural machinery and agricultural robots. Complete the operation tasks of each plot according to the machine instructions. Finally, save the planting data and decision information to the database.

3. The cultivation management and agricultural machinery control method based on large model multi-Agent and Agentic RAG as described in claim 2, wherein the construction of a task planning general agent, understanding and analyzing the data holding status, and generating scheduling instructions, adopts a variant of the supervisor architecture. The task planning general agent acts as the supervisor in the multi-Agent system, responsible for the unified management of this method, using task triggering agents and data agents as tools, and also includes a decision model; The task triggering agent is used to determine the start and continuation nodes of the planting, management, and harvesting tasks, and to release agricultural machinery and materials needed for the warehouse in advance. In addition, it records the details of the day's work. The process includes: (a) The task planning agent issues task instructions, the task triggering agent uses date tools to determine each node of the crop planting stage, and combines weather forecast tools to determine the future workable dates, and records the key date nodes through data recording tools; (b) After determining the operation date, search the network or historical database using search tools to find out the materials required for this operation and transmit them to the agricultural machinery and material warehouse for preparation; (c) On the day the task is due, the task triggering agent sends a task execution signal to the task planning agent, and the task planning agent plans the tasks for the day and uses data agents and decision models according to the plan. (d) After the daily work is completed, record the work details, including information on task completion, information on tasks not completed, and information on missing materials and agricultural machinery; The data agents include environmental data agents, data query agents, data fitting agents, and predictive decision agents. Among them, the data query agent, data fitting agent, and predictive decision agent, together with the historical decision database and the network database, constitute an Agentic RAG. The environmental data agent is used to correct or complete data when the sensor outputs erroneous or missing data. The process includes: (a) Use outlier detection tools to filter data and determine the part to which the outlier data belongs; (b) If there are abnormalities in the sowing, harvesting, and application of pesticides and fertilizers, the abnormalities should be corrected using the status correction tool; (c) If there are environmental data anomalies, obtain environmental information through weather forecasting tools and correct or supplement the anomalies; (d) If soil data anomalies exist, obtain soil information through inference model tools to correct or complete the anomalies; The data query agent is used to query network or historical data and generate network data information. The process includes: (a) Analyze crop information and calculate growth stages using crop growth stage analysis tools; (b) Use search tools to query information on similar crop stages and environmental conditions in the network or historical data to generate original network data information; (c) Data cleaning and preprocessing of the original network data information are performed using data processing tools to obtain the network data information; The predictive decision agent integrates information from multiple sources to make the next decision. The process includes: (a) Use database retrieval tools to retrieve recent environmental and soil information and extract data. (b) Integrate data information with online information through data integration tools; (c) Make decisions based on integrated data using decision-making tools; The data fitting agent is used to generate fitted or predicted data using prompt words when encountering sudden events, and then inputs this data into the decision-making model to play a preventive or remedial role, thereby improving the robustness of the decision-making process. The process includes: (a) Using a prompt generation tool, generate prompts that fit the current environment based on decision information; (b) Input the prompt words into the LLM using a text processing tool and process the resulting LLM output text; (c) Process the obtained text data using data generation tools to obtain a data format suitable for the decision-making model.

4. The cultivation management and agricultural machinery control method based on large-scale multi-agent and agentic RAG according to claim 2, characterized in that: The mechanism-based decision-making model, based on knowledge graph structure and associated knowledge, provides recommendations for agricultural machinery decisions and management methods based on input data. The input data covers the entire planting, cultivation, management, and harvesting process, including environmental and soil information for the planting stage, and crop information added for the cultivation and harvesting stages. The decision-making model will also dynamically adjust according to different operation stages, including activating a multimodal crop disease and pest model during the cultivation and harvesting stages to process crop information. The process includes: (a) Based on the acquired data, generate a list of tasks that need to be performed on the current plot and set priorities. When there is missing information or when prediction is required, this information is generated by the environmental data agent or the prediction decision agent. (b) Generate a list of recommended agricultural machinery based on the task list of the plots, and set priorities based on the functions of the agricultural machinery and fuel consumption information; (c) Select agricultural machinery and materials based on the obtained information and the list of agricultural machinery; (d) When the allocation of agricultural machinery is successful, the equipment and materials to be carried, as well as the recommended operating speed and pesticide application rate, are given, and the operation time of the plot is generated based on the plot area and operating speed. When the allocation of agricultural machinery fails, the plot enters the waiting stage and is redistributed when other agricultural machinery has completed its operation.

5. The cultivation management and agricultural machinery control method based on large-scale multi-agent and agentic RAG according to claim 2, characterized in that: The final decision-making results for agricultural machinery are interacted with in a natural language manner, and instructions are generated to complete the scheduling of agricultural machinery. The process includes: (a) Present agricultural machinery decision-making information in natural language, including recommendations on the amount of machinery, materials, fuel, operating range indicators, and pesticide and fertilizer application standards. (b) Convert natural language into machine instructions that can be recognized by agricultural machinery and agricultural robots, and complete the task of allocating agricultural machinery to each plot according to the machine instructions; (c) Input the planting data and decision-making information after the task is completed into the database.

Images