Smart farm operating system based on agent

Abstract

The invention relates to the technical field of farm management, and discloses an agent-based smart farm operating system, which comprises a resource layer and a business layer, and is characterized in that the resource layer is integrated with a plurality of sensing components, the plurality of sensing components are used for collecting crop data, and the resource layer is also provided with a uniform interface, the business layer is connected with at least one agricultural machine through the uniform interface and comprises a core control module and a plurality of agent modules; wherein the plurality of agent modules are all in signal connection with the resource layer, the plurality of agent modules are mutually connected through a publish-subscribe message bus, and the core control module is connected with the publish-subscribe message bus; and a crop growth model and a task arrangement model are configured in the core control module. The intelligent farm provided by the invention provides an efficient, economic and easy-to-use operation mode, and makes sufficient preparation for future technical progress while meeting current agricultural requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of farm management, and in particular to an agent-based smart farm operating system. Background Art

[0002] Smart farms are an agricultural management model that integrates modern information technology. This advanced agricultural management model not only optimizes all aspects of agricultural production, but also significantly improves productivity and resource utilization efficiency, and vigorously promotes environmental sustainability. In smart farms, farmers can rely on real-time data and automation technology to accurately control key agricultural operations such as irrigation, fertilization and weeding to adapt to the ever-changing natural and market conditions. According to research, these intelligent optimization measures can significantly reduce the use of fertilizers by up to 30%, save water by up to 50%, and have the potential to increase crop yields by more than 20%. The popularization of smart farms will make an important contribution to global food security and sustainable agricultural development.

[0003] At present, the practice of smart farms has integrated many intelligent technologies. For example, the application of Internet of Things (IoT) and remote sensing technology has made data collection more comprehensive and accurate, covering every corner of the farm. The introduction of big data analysis and artificial intelligence technology provides strong support for processing and interpreting these huge data sets, revealing deep information on soil health, weather conditions and crop growth trends. In addition, automation technologies, such as assisted driving tractors and intelligent sprayers, have begun to assist or even replace traditional manual tasks on farms, greatly improving work efficiency and accuracy.

[0004] Despite these exciting advances, smart farms still face many challenges in promoting their widespread adoption and large-scale implementation. First, the use of these artificial intelligence (AI) technologies is currently almost all single-point applications (such as judging pests and diseases), which has failed to form the intelligence of the entire farming operation. Secondly, the management systems of existing farms are basically customized and developed according to the needs of the farms themselves. They are highly complex, have high initial investment costs, and cannot be deployed and migrated quickly. In addition, the operation of these systems requires users to have certain information technology knowledge and operating skills, which is not friendly to farmers. These have restricted the deployment and promotion of smart farms.

[0005] Existing smart farm management systems are usually customized systems, which means that each farm needs to develop a unique solution based on its specific needs. Since farms vary significantly in size, crop types, soil types, and climatic conditions, system developers must tailor a set of software for each farm to suit its specific needs. At present, the bottom layer of the smart farm management system is the resource layer (also known as the data layer), which aggregates various farm-related data. Based on this data, various application logics are developed on the upper layer according to the farm's own needs, and finally displayed through the user interface (UI). This customized development not only increases development costs and time, but also limits the scalability and flexibility of the system. Whenever the needs of the farm change or new agricultural technologies need to be introduced, the system may need to be redesigned and developed, which brings additional burdens to farm managers. In addition, customized development also means that the maintenance and upgrade of the system also requires professional technical support, which hinders the rapid promotion of smart farms.

[0006] At the same time, another significant disadvantage is that most of the existing agricultural management systems are people-centered, focusing on data collection and display. They present the data of various sensors and equipment in the farm in the form of charts and reports for people to make decisions. This people-centered decision-making model requires farm managers to personally analyze a large amount of data and make decisions based on it, which is not only inefficient but also requires high professional skills of managers. For many farmers, these complex data displays are not user-friendly, and they may lack the ability to interpret and apply these data. In addition, this system that relies on manual decision-making has obvious deficiencies in real-time response and automated control, and cannot provide necessary warnings and suggestions in a timely manner, causing farm managers to miss the best management opportunities. Therefore, there is still a lot of room for improvement in the existing system in improving agricultural production efficiency and promoting decision-making automation. Summary of the invention

[0007] The purpose of the present invention is to provide an agent-based smart farm operating system to improve agricultural production efficiency and promote decision-making automation.

[0008] In order to achieve the above purpose, the technical solutions adopted are as follows: The present invention provides an agent-based smart farm operating system, characterized in that it includes a resource layer and a business layer; wherein the resource layer integrates a plurality of sensor components, the plurality of sensor components are used to collect crop data, a unified interface is further provided in the resource layer, at least one agricultural machine is connected through the unified interface, the resource layer is used to obtain the crop data and the operation data fed by the agricultural machine, the crop data and the operation data are stored in the resource layer in the form of set data, the set data form is expressed as (key, value, confidence), wherein key represents the type of data, value represents the data value, confidence represents the accuracy of the data value, if the data comes from the output of a neural network, confidence is a probability value, if the data comes from an actually observed data entry, confidence=1; The business layer includes a core control module and multiple agent modules; wherein the multiple agent modules are all connected to the resource layer signal, the multiple agent modules are interconnected through a publish-subscribe message bus, and the core control module is connected to the publish-subscribe message bus; the core control module is configured with a crop growth model and a task scheduling model, the crop growth model is used to predict the crop growth status based on crop data, when the predicted crop growth status reaches a set stage or develops in an unexpected direction, the agricultural operation task is output to the task scheduling model, the task scheduling model responds to the input agricultural operation task, uses a large language model to decompose the task, and arranges it into an agent execution sequence, based on the agent execution sequence, activates the corresponding agent module to execute the set agricultural task.

[0009] Preferably, in the above-mentioned agent-based smart farm operating system, the agent module has three states, namely, active state, dormant state and inactive state.

[0010] Preferably, in the above-mentioned agent-based smart farm operating system, when the agent module is in an active state, the agent module executes the set agricultural task based on the output of the task scheduling model, and when the set agricultural task is completed or an interrupt instruction issued by the core control module is obtained, the agent module changes from the active state to the dormant state; When the agent module is in a dormant state, the agent module obtains crop data and operation data from the resource layer and feeds them to the core control module. When the agent module obtains a call instruction from other agent modules or an agent execution sequence output by the task scheduling model, the agent module changes from a dormant state to an active state. When the agent module obtains an uninstall instruction, the agent module changes from a dormant state to an inactive state. When the proxy module is in an inactive state, the proxy module is completely stopped and does not perform any tasks. When a manual start instruction is obtained, the proxy module changes from the inactive state to the dormant state.

[0011] Preferably, in the above-mentioned agent-based smart farm operating system, the crop growth model is trained by simulation data and real data collected from multiple farms, and the crop growth model responds to predictions based on a wide range of input data, and outputs crop growth prediction data as predicted growth conditions; wherein the input data includes weather data, soil data, crop management data, crop type and irrigation and fertilization data, and the crop growth prediction data includes future biomass and leaf area index.

[0012] Preferably, in the above-mentioned agent-based smart farm operating system, the crop growth model is a SSAT model, a GECROS model, a WheatSM model or a data-driven AI large model.

[0013] Preferably, in the above-mentioned agent-based smart farm operating system, the multiple agent modules are installed on the core control module in the form of plug-ins; wherein each agent module has a unified interface and a standardized input / output format.

[0014] Preferably, the above-mentioned agent-based smart farm operating system also includes a user interface, and the user interface is communicatively connected to the business layer.

[0015] Preferably, in the above-mentioned agent-based smart farm operating system, the task scheduling model responds to the input agricultural operation task, uses a large language model to decompose the task to obtain at least two subtasks, and schedules the subtasks executed by the agent module in the at least two subtasks into an agent execution sequence, and the at least two subtasks include a soil information collection task, a weather data monitoring task, a sowing condition detection task, an agricultural machinery allocation task, a sowing trajectory planning task, an actual sowing operation task, and a watering and growth monitoring task.

[0016] Preferably, in the above-mentioned agent-based smart farm operating system, the farming operation tasks include sowing tasks and harvesting tasks.

[0017] Preferably, in the above-mentioned agent-based smart farm operating system, after acquiring the crop data and the operation data fed by the agricultural machinery, the resource layer aligns, completes and standardizes the crop data and the operation data and stores them in the resource layer in the form of set data.

[0018] The beneficial effects of the present invention are: (1) Architecture design: The present invention divides the resource layer and the business layer, and designs the core control terminal and the agent module of the business layer, so that the entire system shields the hardware differences downward, allowing the business layer to focus on the business itself; the core control terminal and the agent module are separated, and the core control terminal focuses on the common functions of the farm, thereby realizing convenient migration of different farms, and the agent module realizes basic functions (rather than the entire agricultural operation task), so that different farm needs can be realized by combining different agents, thereby improving the versatility and development cost of the agent.

[0019] (2) Agent mechanism: The use of agent modules in the business layer, which perform specific agricultural functions as independent, pluggable modules. The design and implementation of these agent modules, as well as the communication and coordination mechanisms between them, are applied to the smart farm operating system for the first time.

[0020] (3) Core model: The core control terminal integrates the crop growth model and the task scheduling model to trigger agricultural tasks based on crop growth status prediction. The task scheduling model is used to schedule the execution order of the agents, thus realizing the automation process of the entire agricultural operation from detection to execution.

[0021] (4) Data structure design: Introducing uncertainty into data structure design by adding a confidence field to traditional data tuples to represent the accuracy or probability of the data, thereby enhancing the system's reasoning ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The figure shows a traditional smart farm system architecture diagram according to the prior art.

[0023] Figure 2 An architecture diagram of an agent-based smart farm operating system according to an embodiment of the present invention is shown; in the figure, 100 is a resource layer, 200 is a business layer, 201 is a core control module, 202 is an agent module, 203 is a publish-subscribe message bus, 300 is a sensor component, 400 is an agricultural machinery, and 500 is a user interface.

[0024] Figure 3 A schematic diagram of state transition of an agent module according to an embodiment of the present invention is shown.

[0025] Figure 4 A flow chart of dirt detection of a calibration plate according to an embodiment of the present invention is shown.

[0026] Figure 5 A complete workflow diagram of an agent-based smart farm operating system executing a farming task according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0027] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0028] The specific implementation of the present invention is further described in detail below in conjunction with the drawings and examples.

[0029] Figure 1 This is a diagram of the traditional smart farm system architecture. Figure 1 As shown in the figure, the traditional smart farm system architecture includes a resource layer and a business layer. The resource layer is connected to sensors and agricultural equipment, and the business layer is equipped with multiple business logics. In this traditional smart farm system architecture, system components are highly customized and tightly coupled, making it difficult to copy and migrate between different farms. Each system must be rebuilt according to specific environmental conditions and operating requirements, resulting in a significant increase in development costs.

[0030] Based on this, the embodiment of the present invention provides an agent-based smart farm operating system, which automates and intelligentizes the execution of agricultural tasks through modular agents and artificial intelligence-driven orchestration. The agent-based smart farm operating system takes crops as the core, monitors the growth of crops, discovers, makes decisions and processes abnormal growth conditions, thereby realizing the collection of decision support. People only exist for the better progress of the entire process, which improves the efficiency of agricultural production and the sustainability of resource utilization. Compared with the traditional system, the agent-based smart farm operating system decomposes tasks into flexibly configurable agents, each agent implements some basic functions in agricultural operations, and for each specific agricultural operation, only the appropriate agent needs to be selected for orchestration. The agent-based smart farm operating system provides management of the operation of each agent module. This method not only meets the functional requirements of different farms, but also unifies the implementation of the agent-based smart farm operating system, simplifies farm management, reduces dependence on manual labor, and reduces system migration costs. In addition, the design of the agent-based smart farm operating system allows the system to adapt to the development of future technologies, providing a solid foundation for the long-term development of smart agriculture. In general, the agent-based smart farm operating system provides an efficient, economical and easy-to-use operation mode for smart farms, which not only meets current agricultural needs but also prepares for future technological advances.

[0031] Specifically, Figure 2 The following is an architecture diagram of an agent-based smart farm operating system provided by an embodiment of the present invention. Figure 2 As shown, the agent-based smart farm operating system includes a resource layer 100 and a business layer 200; wherein, the resource layer 100 is integrated with multiple sensor components 300, and the multiple sensor components 300 are used to collect crop data. A unified interface is also provided in the resource layer 100, and at least one agricultural machine 400 is connected through the unified interface. The resource layer 100 is used to obtain crop data and operation data fed by the agricultural machine. The crop data and the operation data are stored in the resource layer 100 in the form of set data. The set data form is expressed as (key, value, confidence), wherein key represents the type of data, value represents the data value, and confidence represents the accuracy of the data value. If the data comes from the output of the neural network, confidence is a probability value. If the data comes from the actual observed data entry, confidence=1.

[0032] As a solid foundation of the agent-based smart farm operating system, the resource layer 100 is responsible for the centralized management and operation of sensors and agricultural machinery in the farm. By integrating a variety of sensors as sensing components, the resource layer collects key data on soil conditions, climate conditions, and crop growth, while providing a unified interface for various agricultural machinery to achieve centralized scheduling and control.

[0033] In order to support uncertainty-based reasoning, the agent-based smart farm operating system introduces the concept of uncertainty in the data structure design of the resource layer 100. Specifically, the traditional data tuple (key, value) is expanded to (key, value, confidence), where key represents the type of data, value represents the data value, and confidence represents the accuracy level of the data value. If the data value comes from the output of a neural network, then confidence represents the corresponding probability. For data entries observed by sensors or existing traditional systems, if they enter the agent-based smart farm operating system in the form of (key, value) (i.e., without confidence components), these entries will be expanded to the form of (key, value, 1). This design allows the system to take into account the uncertainty of the data when processing data, thereby improving the accuracy and reliability of decision-making. Through this carefully designed data structure, the resource layer can not only efficiently collect and process a large amount of agricultural data, but also make effective reasoning and decisions when the uncertainty of the data exists. This built-in support for uncertainty enables the agent-based smart farm operating system to make more accurate and reliable decisions when facing complex and changing agricultural environments, thereby improving the efficiency and effectiveness of agricultural production. At the same time, the design of this data structure also reflects the advancement and innovation of the agent-based smart farm operating system in processing agricultural data, providing strong data support for the development of smart agriculture.

[0034] It should be noted that the data items observed by the sensors or existing traditional systems mentioned above, for example, can be the ambient temperature data collected by the temperature sensor, and the data form of the ambient temperature data can be (ambient temperature, 25°C, 1), that is, the value of confidence is assigned to 1. In the case where the data value comes from the output of the neural network, such as a visual sensing module, the visual sensing module usually includes a camera and a neural network for image recognition mounted in the camera, and the neural network is used to identify the crop type in the image based on the image captured by the camera. Then, the data collected by the visual sensing module in the resource layer 100 is in the form of (crop type, corn, 0.95), and confidence = 0.95 indicates the probability that the neural network predicts that the crop type is corn. It should be noted that value can be a field or a specific value according to different keys. The data form obtained after the data collected by the two sensor components exemplified above and processed by the resource layer 100 is only an example and does not constitute a limitation to the present invention.

[0035] In an exemplary embodiment, after acquiring the crop data and the operation data fed by the agricultural machinery, the resource layer aligns, completes and standardizes the crop data and the operation data and stores them in the resource layer in the form of set data.

[0036] The data of the resource layer 100 is precisely aligned in the time and space dimensions, providing accurate and standardized information input for the business logic of the business layer 200, effectively shielding the upper-layer application from the perception of hardware differences and ensuring the flexibility and scalability of the system. This design makes the resource layer a powerful data collection and processing center, providing a solid foundation for the intelligent management of the entire farm.

[0037] The business layer 200 includes a core control module 201 and multiple agent modules 202; wherein, the multiple agent modules 202 are all connected to the resource layer 100 by signal, the multiple agent modules 202 are interconnected through a publish-subscribe message bus 203, and the core control module 201 is connected to the publish-subscribe message bus 203; the core control module 201 is configured with a crop growth model and a task scheduling model, the crop growth model is used to predict the crop growth status based on crop data, when the predicted crop growth status reaches a set stage or develops in an unexpected direction, the agricultural operation task is output to the task scheduling model, the task scheduling model responds to the input agricultural operation task, uses a large language model to decompose the task, and arranges it into an agent execution sequence, based on the agent execution sequence, activates the corresponding agent module 202 to execute the set agricultural task.

[0038] The business layer 200 is built on the resource layer 100 and is the core of the agent-based smart farm operating system. It consists of a series of agent modules 202 that perform specific functions, an efficient publish-subscribe message bus 203, and a core control module 201 configured with two core models. Each agent module 202 communicates and coordinates through the publish-subscribe message bus 203, so that it can listen to specific messages and be activated accordingly, thereby achieving rapid response and precise execution of agricultural tasks. The agent module 202 is the basic component of the agent-based smart farm operating system. Each agent module 202 obtains information from the resource layer 100, completes its designated function, and returns the output to the resource layer 100. Beehive provides general agent management and orchestration capabilities, effectively coordinating the relationship between the system's versatility and the farm's customization needs.

[0039] In an exemplary embodiment, the agent-based smart farm operating system further includes a user interface 500. The user interface 500 is in communication with the business layer 200 and is used to display the data acquired by the resource layer 100 and the processed intermediate data and processing structure data of the business layer 200.

[0040] In an exemplary embodiment, the agent module has three states, namely, an active state, a dormant state, and an inactive state.

[0041] In this embodiment, management of the agent module 202 is provided in the agent-based smart farm operating system. This management method is independent of farm business, so different farms can use the same management method. The state transition of the agent module 202 is a key mechanism for realizing task automation. The agent module 202 switches between three states: active state, dormant state, and inactive state: in the active state, it executes the specified task; in the dormant state, it waits for activation but can still respond to specific signals; in the inactive state, it stops completely until manually restarted.

[0042] In one embodiment, if Figure 3 , which is a schematic diagram of the state transition of the agent module provided by an embodiment of the present invention. The agent module 202 can be activated by directly calling the task scheduling model or by subscribing to messages on the message bus. Once the task is completed or interrupted, it returns to the sleep state. Before uninstalling the agent module 202, it must be ensured that it is in the sleep state. This state management ensures the flexibility and responsiveness of the system, so that the agent-based smart farm operating system can efficiently manage and schedule agricultural tasks.

[0043] Specifically, when the agent module 202 is in an active state, the agent module 202 executes the set agricultural task based on the output of the task scheduling model. When the set agricultural task is completed or an interrupt instruction issued by the core control module 201 is obtained, the agent module 202 changes from the active state to the dormant state; When the agent module 202 is in a dormant state, the agent module 202 obtains crop data and operation data from the resource layer 100 and feeds them to the core control module 201. When obtaining a call instruction from other agent modules 202 or an agent execution sequence output by the task scheduling model, the agent module 202 changes from the dormant state to the active state. When obtaining an uninstall instruction, the agent module 202 changes from the dormant state to the inactive state. When the agent module 202 is in an inactive state, the agent module 202 is completely stopped and does not execute any task. When a manual start instruction is obtained, the agent module 202 changes from the inactive state to the dormant state.

[0044] In an exemplary embodiment, a plurality of proxy modules 202 are installed on the core control module in the form of plug-ins; wherein each proxy module has a unified interface and a standardized input / output format.

[0045] In this embodiment, in order to standardize management and scheduling, the agent module 202 is designed to have a unified interface and standardized input / output format. The agent module 202 is installed on the business layer 200 as a plug-in, providing various basic functions in the entire agricultural operation process of cultivation, management and harvesting, and most of the functions need to be implemented using artificial intelligence technology. According to the actual needs of the farm, an agent with corresponding functions can be selected or developed. This design decomposes the complex needs of the farm and turns it into on-demand installation and organization of the agent, which simplifies complex problems and makes them easy to implement. These agent modules 202 can be developed by a team of developers with different backgrounds. The core model is responsible for aligning and converting the input and output of the connected agents during the orchestration process to solve the problem that when representing the same information, the data required by the agent module 202 and the data provided by the system may differ. Compared with the traditional system, the newly added agent module 202 design allows the system to flexibly select or develop agents with corresponding functions according to the specific needs of the farm, realizing the modularization and on-demand configuration of the complex needs of the farm. This modular approach not only simplifies complex problems, but also enables the system to easily adapt to the unique needs of different farms.

[0046] In an exemplary embodiment, a crop growth model is trained by simulation data and real data collected from multiple farms, and the crop growth model responds to predictions based on a wide range of input data and outputs crop growth prediction data as predicted growth conditions; wherein the input data includes weather data, soil data, crop management data, crop type and irrigation and fertilization data, and the crop growth prediction data includes future biomass and leaf area index.

[0047] Exemplarily, the crop growth model may be at least one of a SSAT model, a GECROS model, a WheatSM model, and a data-driven AI big model.

[0048] In an exemplary embodiment, the task scheduling model responds to an input agricultural operation task, uses a large language model to decompose the task to obtain at least two subtasks, and schedules the subtasks executed by the agent module in the at least two subtasks into an agent execution sequence, and the at least two subtasks include a soil information collection task, a weather data monitoring task, a sowing condition detection task, an agricultural machinery allocation task, a sowing trajectory planning task, an actual sowing operation task, and a watering and growth monitoring task.

[0049] The crop growth model and the task orchestration model are two core models, which are configured in the core control module 201 and serve as the brain of the entire agent-based smart farm operating system. The crop growth model focuses on monitoring and predicting the growth status of crops. It can be a simple traditional mechanism model such as DSSAT, GECROS and WheatSM, or a data-driven AI model that uses these traditional mechanism models to generate a large amount of simulation data and real-world data collected from multiple farms for training. This model can make predictions based on a wide range of input data such as weather, soil, crop management, crop type and irrigation and fertilization, and output crop growth prediction data for future periods such as biomass and leaf area index (LAI). With the help of the Transformer architecture and sufficient training data, the model shows strong generalization capabilities and can adapt to different crops and geographical environments. This capability is lacking in traditional systems, which are usually unable to make effective predictions and adaptations across different conditions and environments. The task orchestration model makes decisions based on the output of the crop growth model and available resources, and uses a large language model (LLM) to orchestrate agents to execute these decisions.

[0050] like Figure 4 As shown in the figure, for the "corn planting" operation, the task orchestration model decomposes the operation into multiple functions and orchestrates the order of agent execution of the corresponding functions. This model takes over the decision-making role traditionally held by farmers or experts, initially providing basic functions and keeping people-centric. As the model's capabilities develop, it will gradually automate more decisions and assist farmers in fully automating most processes. A few key decisions still require manual input, but they can gradually develop into a fully autonomous task execution system. This level of automation and intelligence is not available in traditional agricultural systems, which often rely on manual decision-making and intervention and lack efficient and intelligent task processing capabilities.

[0051] Figure 4 A detailed process of corn sowing is shown, starting from the output of sowing suggestions by the large language model (LLM) to the various steps of completing the sowing task through agent coordination, such as soil information collection, weather data monitoring, sowing condition detection, agricultural machinery allocation, sowing trajectory planning, actual sowing operation, and subsequent watering and growth monitoring. The entire process reflects the intelligent capabilities of the Beehive system in task scheduling and agent activation, and how to improve agricultural production efficiency and accuracy through automation methods.

[0052] like Figure 5As shown, it is a complete workflow diagram of an agent-based smart farm operating system provided in an embodiment of the present invention to perform an agricultural operation task. The crop growth model predicts the growth status of crops based on the data collected and organized by the resource layer 100. When it is predicted that the crop growth reaches a certain stage (such as reaching a certain agricultural stage) or develops in an unexpected direction (such as poor growth), the corresponding agricultural operations (such as sowing, harvesting, etc.) are triggered. The task scheduling model decomposes the task based on the operation task and existing resources (such as the number and type of agricultural machinery, etc.) (using a large language model), and arranges it into a corresponding agent execution sequence. The agent module 202 is activated and performs a specific agricultural task. After the execution is completed, it represents the completion of the entire agricultural operation. Multiple agricultural operations can be performed in parallel.

[0053] The present invention makes it more efficient and accurate in handling agricultural tasks through the design of two core models. Compared with traditional systems that require a lot of manual operation and monitoring, the system proposed by the present invention significantly improves the degree of automation of agricultural production through advanced agent management and intelligent decision support. For example, the task scheduling model can automatically decompose complex agricultural tasks and execute these tasks in parallel or sequentially by activating relevant agents, thereby optimizing resource allocation and job scheduling. This capability not only improves the efficiency of farm operations, but also enhances adaptability and response speed to environmental changes, ultimately improving overall agricultural productivity and sustainability.

[0054] In summary, compared with Figure 1 Compared with the traditional smart farm system shown in the figure, the agent-based smart farm operating system proposed in this embodiment, through a layered architecture and agent mechanism, not only provides a highly adaptable and scalable automation solution, but also greatly reduces the complexity of management and significantly reduces the reliance on manual supervision. At the same time, the agent-based smart farm operating system takes uncertainty into consideration through its advanced data structure design, enabling the system to more accurately process and reason about data from sensors and agricultural machinery. Combined with the powerful computing power of artificial intelligence and large language models, the agent-based smart farm operating system also realizes full-process automation from real-time data collection to decision support, greatly improving the efficiency of agricultural production and the sustainability of resources.

[0055] The design flexibility of the agent-based smart farm operating system allows it to adapt to changing farm conditions and operating requirements, whether it is different climate environments, soil types or crop types, it can be intelligently adapted. This is particularly prominent during system migration and development, because the modular design of the agent-based smart farm operating system greatly reduces migration costs and development time. In addition, by reducing reliance on manual intervention, the system greatly reduces the user's learning threshold and operating difficulty, allowing even less technically skilled users, such as aging agricultural labor, to easily manage and operate. This user-friendly design concept makes the agent-based smart farm operating system suitable not only for large farms, but also for small farms and family farms, greatly expanding the coverage and applicability of smart agricultural technology.

[0056] The above implementation modes are only used to illustrate the present invention, but not to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, all equivalent technical solutions also belong to the scope of the present invention. The patent protection scope of the present invention should be defined by the claims.

Claims

1. An agent-based smart farm operating system, characterized in that: It includes a resource layer and a business layer; wherein the resource layer integrates a plurality of sensor components, the plurality of sensor components are used to collect crop data, the resource layer is also provided with a unified interface, at least one agricultural machine is connected through the unified interface, the resource layer is used to obtain the crop data and the operation data fed by the agricultural machine, the crop data and the operation data are stored in the resource layer in the form of set data, the set data form is expressed as (key, value, confidence), wherein key indicates the type of data, value indicates the data value, confidence indicates the accuracy of the data value, if the data comes from the output of the neural network, confidence is a probability value, if the data comes from the actually observed data entry, confidence=1; The business layer includes a core control module and multiple agent modules; wherein the multiple agent modules are all connected to the resource layer signal, the multiple agent modules are interconnected through a publish-subscribe message bus, and the core control module is connected to the publish-subscribe message bus; the core control module is configured with a crop growth model and a task scheduling model, the crop growth model is used to predict the crop growth status based on crop data, when the predicted crop growth status reaches a set stage or develops in an unexpected direction, the agricultural operation task is output to the task scheduling model, the task scheduling model responds to the input agricultural operation task, uses a large language model to decompose the task, and arranges it into an agent execution sequence, based on the agent execution sequence, activates the corresponding agent module to execute the set agricultural task.

2. The agent-based smart farm operating system according to claim 1, characterized in that: The proxy module has three states, namely, an active state, a dormant state and an inactive state.

3. The agent-based smart farm operating system according to claim 2, characterized in that: When the proxy module is in an active state, the proxy module executes the set agricultural task based on the output of the task scheduling model, and when the set agricultural task is completed or an interrupt instruction issued by the core control module is obtained, the proxy module changes from the active state to the dormant state; When the agent module is in a dormant state, the agent module obtains crop data and operation data from the resource layer and feeds them to the core control module. When the agent module obtains a call instruction from other agent modules or an agent execution sequence output by the task scheduling model, the agent module changes from a dormant state to an active state. When the agent module obtains an uninstall instruction, the agent module changes from a dormant state to an inactive state. When the proxy module is in an inactive state, the proxy module is completely stopped and does not perform any tasks. When a manual start instruction is obtained, the proxy module changes from the inactive state to the dormant state.

4. The agent-based smart farm operating system according to claim 1, characterized in that: The crop growth model is trained by simulation data and real data collected from multiple farms. The crop growth model responds to predictions based on a wide range of input data and outputs crop growth prediction data as predicted growth conditions; wherein the input data includes weather data, soil data, crop management data, crop type and irrigation and fertilization data, and the crop growth prediction data includes future biomass and leaf area index.

5. The agent-based smart farm operating system according to claim 1, characterized in that: The crop growth model is a SSAT model, a GECROS model, a WheatSM model or a data-driven AI large model.

6. The agent-based smart farm operating system according to claim 1, characterized in that: The multiple proxy modules are installed on the core control module in the form of plug-ins; wherein each proxy module has a unified interface and a standardized input / output format.

7. The agent-based smart farm operating system according to claim 1, characterized in that: It also includes a user interface, which is communicatively connected with the business layer.

8. The agent-based smart farm operating system according to claim 1, characterized in that: The task scheduling model responds to the input agricultural operation task, uses a large language model to decompose the task to obtain at least two subtasks, and schedules the subtasks executed by the agent module in the at least two subtasks into an agent execution sequence, and the at least two subtasks include a soil information collection task, a weather data monitoring task, a sowing condition detection task, an agricultural machinery allocation task, a sowing trajectory planning task, an actual sowing operation task, and a watering and growth monitoring task.

9. The agent-based smart farm operating system according to claim 8, characterized in that: The agricultural operation tasks include sowing tasks and harvesting tasks.

10. The agent-based smart farm operating system according to claim 1, characterized in that: After acquiring the crop data and the operation data fed by the agricultural machinery, the resource layer aligns, completes and standardizes the crop data and the operation data and stores them in the resource layer in the form of set data.

Images