Information decision-making methods, devices, systems, electronic devices, storage media and products

Abstract

This application relates to the fields of artificial intelligence and data processing technology, providing an information decision-making method, apparatus, system, electronic device, storage medium, and product. The information decision-making method is applied to a blackboard sharing network. The method includes: receiving decision-based data sent by a perceptual agent and publishing the decision-based data; receiving an intent reasoning conclusion fed back by an intent reasoning agent based on the decision-based data and publishing the intent reasoning conclusion; and receiving a task orchestration result fed back by a task orchestration agent based on the decision-based data and the intent reasoning conclusion. This application constructs an asynchronous, decoupled, data-driven information decision-making method based on a blackboard sharing network, enabling agents with different responsibilities to independently and autonomously conduct adaptive collaboration, efficiently and flexibly responding to complex and dynamic task requirements, and significantly improving the flexibility and execution efficiency of complex information decision-making task orchestration.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and data processing technology, specifically to an information decision-making method, apparatus, system, electronic device, storage medium, and product. Background Technology

[0002] Information-based decision-making occupies a central position in all sectors of modern society. It refers to the process of systematically collecting, analyzing, and applying relevant data and knowledge to optimize action plans, reduce uncertainty, and improve the quality of outcomes. In a complex and ever-changing environment, the quality of information-based decision-making directly determines the efficiency and effectiveness of actions—it helps mitigate risks, reduce resource waste, capture potential opportunities, and drive systemic innovation. Whether in business management, public policy formulation, or daily life choices, accurate information-based decision-making is the foundation for achieving rational action and sustainable development.

[0003] Taking agricultural information decision-making as an example, agricultural production is highly dependent on the natural environment. Crop growth involves multiple variables such as climate, soil, water, and pests and diseases, and is highly regional and seasonal. Traditional experience-based decision-making methods often struggle to take a holistic approach and anticipate risks. However, agricultural information decision-making systems can integrate and analyze information such as seasonal changes, climate conditions, and land resources, enabling dynamic perception and intelligent analysis of farmland conditions. Furthermore, they can rationally arrange and organize agricultural production activities according to the crop growth cycle, thereby helping to improve crop yield and quality.

[0004] Currently, there are two main technical solutions for automating the modeling and organization of information elements in various information decision-making scenarios through information systems and large models: (1) based on flowcharts / custom orchestration engines, and (2) based on agent frameworks for orchestration. Although these two technical solutions can involve multiple agents collaborating, the collaboration between agents is executed in a human-defined order. For a given task, the sub-tasks are first decomposed by the thought chain, then the sub-tasks are distributed, and finally the results are summarized. For new tasks that are not predetermined, it is impossible for agents to completely independently and adaptively collaborate, resulting in poor adaptability to complex tasks.

[0005] Therefore, how to achieve adaptive collaboration among multiple agents to improve the flexibility and efficiency of information decision-making task orchestration is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] This application provides an information decision-making method, apparatus, system, electronic device, storage medium, and product. By constructing an asynchronous, decoupled, data-driven information decision-making method based on a blackboard sharing network, it enables intelligent agents with different responsibilities to independently and autonomously conduct adaptive collaboration, efficiently and flexibly respond to complex and dynamic task requirements, and significantly improve the flexibility and execution efficiency of complex information decision-making task orchestration.

[0007] In a first aspect, embodiments of this application provide an information decision-making method applied to a blackboard sharing network, comprising: Receive basic decision-making data sent by the perceptual agent, and publish the basic decision-making data; Receive the intent reasoning conclusion fed back by the intent reasoning agent based on the decision-making basis data, and publish the intent reasoning conclusion; The task orchestration agent receives the task orchestration results based on the decision-making data and the intent reasoning conclusion.

[0008] In one embodiment, the information decision-making method further includes: When a conflict is detected between the intent reasoning conclusion and the task orchestration result, a conflict event is issued; Receive the arbitration result corresponding to the conflict event sent by the decision arbitration agent, the arbitration result including the winning agent and the lock time; Based on the arbitration result, a lock announcement is issued. The lock announcement is used to indicate that the content of the winning agent is valid during the lock period and to instruct the unsuccessful agents to enter a backoff waiting state.

[0009] In one embodiment, the arbitration result is determined in the following manner: Based on the conflict event, obtain the target conflict agent and its current weight, the rationality score of the conflict content, and the time difference between the occurrence time of the conflict event and the current time; Based on the current weight, the rationality score, and the time difference, calculate the arbitration priority score for each of the target conflict agents; The agent with the highest arbitration priority score is selected as the winning agent, and the locking time is calculated based on the arbitration priority score of the winning agent.

[0010] In one embodiment, after issuing the lock announcement based on the arbitration result, the information decision-making method further includes: Obtain the content feedback score of the target conflict agent and obtain the average feedback score; The receiving feedback learning agent sends updated weights based on the content feedback score, the average feedback score, and the current weights; The updated weights are determined as follows: Calculate the difference between the content feedback score and the average feedback score; Using a reinforcement learning algorithm, the current weights of the winning agent are updated based on the difference.

[0011] In one embodiment, the arbitration result further includes the backoff time of the unsuccessful agent, which is calculated based on the lock time, dynamic delay time, and random perturbation value, wherein the dynamic delay time is determined based on the number of historical failures of the unsuccessful agent. After issuing a lock announcement based on the arbitration result, wherein the lock announcement is used to indicate that the content of the winning agent is valid during the lock period and to instruct the unsuccessful agents to enter a backoff waiting state, the information decision-making method further includes: Receive the resubmission request triggered by the unsuccessful agent after the backoff time has elapsed; If no conflict is detected between the resubmitted requests, the corresponding operation is performed based on the resubmitted requests.

[0012] In one embodiment, before receiving the task orchestration result fed back by the task orchestration agent based on the decision-making foundation data and the intent reasoning conclusion, the information decision-making method further includes: Receive available device information sent by the device control agent, and publish the available device information; The task orchestration result received by the receiving task orchestration agent based on the decision-making data and the intent reasoning conclusion includes: Receive the task orchestration result fed back by the task orchestration agent based on the decision-making data, the intent reasoning conclusion, and the available device information; The task orchestration result is obtained by the task orchestration agent in the following manner: Based on the intent reasoning conclusion, the task to be executed is determined, and the task to be executed is decomposed to obtain the sub-tasks to be executed; Based on the decision-making data and the available device information, the execution priority score of the task to be executed is calculated; Based on the execution priority score, the available device information, and the sub-tasks to be executed, the tasks to be executed are sorted and resources are allocated to obtain the task orchestration result.

[0013] In one embodiment, the information decision-making method further includes: Receive the task execution result and effect verification request fed back by the device control agent based on the task orchestration result, and publish the task execution result and the effect verification request; Receive effect verification data sent by the perceptual agent based on the effect verification request, and publish the effect verification data; The receiving feedback learning agent receives model parameters based on the effect verification data, the decision-making basis data, and the task execution results. These model parameters are used to calculate the arbitration priority score.

[0014] Secondly, embodiments of this application provide an information decision-making device, including: The basic data receiving module is used to receive and publish the basic decision data sent by the perceptual intelligent agent. The reasoning conclusion receiving module is used to receive the intention reasoning conclusion fed back by the intention reasoning agent based on the decision-making basis data, and to publish the intention reasoning conclusion; The orchestration result receiving module is used to receive the task orchestration result fed back by the task orchestration agent based on the decision-making basis data and the intent reasoning conclusion.

[0015] Thirdly, embodiments of this application provide an information decision-making system, including a blackboard sharing network, a perceptual agent, an intent reasoning agent, and a task orchestration agent; wherein: The blackboard sharing network is used to execute the information decision-making method described in the first aspect above; The sensing agent is used to collect basic decision-making data and send the basic decision-making data to the blackboard sharing network; The intent reasoning agent is used to obtain the decision-making basis data from the blackboard sharing network, determine the intent reasoning conclusion based on the decision-making basis data, and feed back the intent reasoning conclusion to the blackboard sharing network. The task orchestration agent is used to obtain the decision-making basis data and the intent reasoning conclusion from the blackboard sharing network, determine the task orchestration result based on the decision-making basis data and the intent reasoning conclusion, and feed back the task orchestration result to the blackboard sharing network.

[0016] In one embodiment, the information decision-making system further includes a decision arbitration intelligent agent; The blackboard sharing network is used to publish a conflict event when a conflict is detected between the intention reasoning conclusion and the task orchestration result; The decision-making arbitration agent is used to determine the arbitration result corresponding to the conflict event and send the arbitration result to the blackboard sharing network; wherein, the arbitration result includes the winning agent and the lock time; The blackboard sharing network is also used to issue a lock announcement based on the arbitration result. The lock announcement is used to indicate that the content of the winning agent is valid during the lock time and to instruct the unsuccessful agents to enter a backoff waiting state.

[0017] In one embodiment, the information decision-making system further includes a device control agent and a feedback learning agent; The device control agent is used to obtain the task orchestration results from the blackboard sharing network, execute the task, and send the task execution results and effect verification requests to the blackboard sharing network. The sensing agent is also used to collect effect verification data and send the effect verification data to the blackboard sharing network when it receives the effect verification request from the blackboard sharing network. The feedback learning agent is used to obtain the decision-making basis data, the task execution results, and the effect verification data from the blackboard sharing network, and to optimize the model parameters using reinforcement learning algorithms.

[0018] Fourthly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the information decision-making method described in the first aspect.

[0019] Fifthly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the information decision-making method described in the first aspect.

[0020] In a sixth aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the information decision-making method described in the first aspect.

[0021] The information decision-making method, apparatus, system, electronic device, storage medium, and product provided in this application are applied to a blackboard sharing network. The blackboard sharing network first receives and publishes basic decision-making data sent by the perceptual agent. This means that the system's decision-making process is driven by real-world data changes rather than preset instructions, solving the dependence on predetermined tasks in existing technologies. This allows the system to adaptively discover and respond to unexpected new tasks in the environment, greatly enhancing its adaptability to complex and dynamic environments. Subsequently, the blackboard sharing network receives and publishes intent reasoning conclusions fed back by the intent reasoning agent based on the aforementioned data, and task orchestration results fed back by the task orchestration agent based on the basic decision-making data and intent reasoning conclusions. Through this approach, an asynchronous, decoupled, data-driven information decision-making method based on a blackboard sharing network is constructed. Each agent, as an independent decision-making unit, independently performs information reading and writing, data perception, reasoning, and planning around the blackboard sharing network, realizing adaptive collaboration among multiple agents. Ultimately, without human intervention, it efficiently and flexibly responds to complex and dynamic task requirements, significantly improving the flexibility and execution efficiency of complex information decision-making task orchestration. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is one of the flowcharts illustrating the information decision-making method provided in the embodiments of this application.

[0024] Figure 2 This is the second flowchart illustrating the information decision-making method provided in the embodiments of this application.

[0025] Figure 3 This is an example operation timing diagram of the information decision-making method provided in the embodiments of this application.

[0026] Figure 4 This is the third flowchart illustrating the information decision-making method provided in the embodiments of this application.

[0027] Figure 5 This is a schematic diagram of the structure of the information decision-making system provided in the embodiments of this application.

[0028] Figure 6 This is a schematic diagram of the information decision-making device provided in the embodiments of this application.

[0029] Figure 7This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] Information-based decision-making occupies a central position in all sectors of modern society. It refers to the process of systematically collecting, analyzing, and applying relevant data and knowledge to optimize action plans, reduce uncertainty, and improve the quality of outcomes. In a complex and ever-changing environment, the quality of information-based decision-making directly determines the efficiency and effectiveness of actions—it helps mitigate risks, reduce resource waste, capture potential opportunities, and drive systemic innovation. Whether in business management, public policy formulation, or daily life choices, accurate information-based decision-making is the foundation for achieving rational action and sustainable development.

[0032] Taking agricultural information decision-making as an example, agricultural production is highly dependent on the natural environment. Crop growth involves multiple variables such as climate, soil, water, and pests and diseases, and is highly regional and seasonal. Traditional experience-based decision-making methods often struggle to take a holistic approach and anticipate risks. However, agricultural information decision-making systems can integrate and analyze information such as seasonal changes, climate conditions, and land resources, enabling dynamic perception and intelligent analysis of farmland conditions. Furthermore, they can rationally arrange and organize agricultural production activities according to the crop growth cycle, thereby helping to improve crop yield and quality.

[0033] In recent years, with the widespread application of technologies such as artificial intelligence, the Internet of Things (IoT), and remote sensing in agriculture, intelligent decision-making systems based on agricultural information have gradually emerged. In modern agriculture, intelligent equipment (such as agricultural drones, intelligent irrigation equipment, and automatic fertilization robots) has initially achieved partial automation control of the planting process. Meanwhile, large-scale models for agricultural scenarios (such as the Shennong Big Model and Gengyun Model) can also provide planting suggestions and crop growth analysis capabilities. Regarding intelligent systems, most adopt centralized scheduling, where the central system analyzes the current state and issues tasks according to fixed rules. Each execution module receives the commands and completes the operation, and feedback information is usually collected manually or uploaded through sensing devices. Currently, there are two main technical solutions for automating the modeling and organization of the above agricultural information elements in agricultural production activities through information systems and large-scale models.

[0034] The first approach is based on a flowchart / custom orchestration engine. This method logically organizes multiple large model capabilities (such as question answering, image recognition, reasoning, and planning) with external tools (querying, retrieval, external knowledge bases, knowledge graphs, etc.), service APIs (weather query services, geolocation query services, remote sensing analysis services, etc.), and task flows (sequential execution, branch execution, loop execution, conditional execution) to form an executable composite application. Each node represents an operation step, and nodes are organized through logical connections. This approach emphasizes task visualization and process control, and is suitable for scenarios with clear task boundaries and well-defined operation steps.

[0035] The second approach involves orchestration based on an agent framework (such as LangChain): multiple agents are used to handle different tasks, with the master agent responsible for decision-making and scheduling. Each agent can call sub-models, APIs, search tools, etc. By supporting multi-round thinking and autonomous planning structures, a composite decision-making system based on agricultural information is ultimately achieved. Agents based on large models have stronger contextual understanding and dynamic decision-making capabilities, making them suitable for complex scenarios with blurred task boundaries and frequent environmental changes.

[0036] While the agricultural information decision-making systems described above can involve multiple agents collaborating, this collaboration occurs in a pre-defined sequence. For a given task, sub-tasks are first decomposed using a thought chain, then distributed, and finally the results are aggregated. For newly added, unpredictable tasks, agents cannot independently and adaptively collaborate, resulting in poor adaptability to complex tasks.

[0037] Therefore, how to achieve adaptive collaboration among multiple agents to improve the flexibility and efficiency of information decision-making task orchestration is a technical problem that urgently needs to be solved in this field.

[0038] Based on the above, this application proposes an information decision-making method, apparatus, system, electronic device, storage medium, and product, which will be discussed below. Figures 1-7 Describe it.

[0039] Figure 1 This is one of the flowcharts illustrating the information decision-making method provided in the embodiments of this application. (Refer to...) Figure 1 This application provides an information decision-making method applied to a blackboard sharing network, including steps S110, S120 and S130.

[0040] Step S110: Receive the decision-making basis data sent by the perceptual agent and publish the decision-making basis data.

[0041] In this embodiment, the information decision-making method is applied to a blackboard sharing network, meaning the blackboard sharing network is the executing entity of the information decision-making method. The information decision-making method of this embodiment can be applied to scenarios such as enterprise management, public policy formulation, daily life choices, healthcare, and agricultural information decision-making. This embodiment uses an agricultural information decision-making scenario as an example for illustration.

[0042] In this embodiment, the information decision-making system includes multiple intelligent agents with clearly defined functions and responsibilities, such as a perception agent, an intent reasoning agent, and a task orchestration agent. Furthermore, a blackboard sharing network is introduced. The blackboard sharing network is the core mechanism for multi-agent collaboration in this information decision-making system. It performs the following three functions: ① Information sharing: All agents can read and write state information on the blackboard, achieving data synchronization; ② Event triggering: When an agent writes new state information, other agents can detect the change and respond automatically; ③ Task coordination: The blackboard records task execution status, feedback results, etc., supporting collaboration and scheduling among multiple agents. Through the blackboard sharing network, the system achieves decentralized agent collaboration. Each agent operates independently and can respond based on the global state, improving the system's flexibility, scalability, and fault tolerance.

[0043] Decision-making data refers to raw, unprocessed data or preliminarily processed data collected by perceptual agents for subsequent decision-making. Taking agricultural information decision-making scenarios as an example, decision-making data may include, but is not limited to: crop growth images, weather data, and soil data.

[0044] A perceptual agent is used to send collected or pre-processed decision-making data to the blackboard sharing network. For example, it can perceive the crop type, growth stage, and environmental status of a plot of land through visual image recognition, remote sensing data analysis, and IoT devices (such as soil moisture sensors, light sensors, etc.), and send the collected or pre-processed decision-making data to the blackboard sharing network. Correspondingly, the blackboard sharing network receives the decision-making data sent by the perceptual agent and publishes the decision-making data. The publishing methods include, but are not limited to: (1) writing to the blackboard data area and making it public to all agents; (2) writing to the blackboard data area and making it publicly available to agents that have subscribed to the decision-making data; (3) writing to the blackboard data area and sending it to agents that have subscribed to the decision-making data.

[0045] Furthermore, when the decision-making basis data is data that has been initially processed by the sensing agent, its raw data can come from the Internet of Things (IoT) infrastructure. The IoT infrastructure includes a series of devices that can be used for agricultural information collection. Then, analog signals are converted into digital signals and uploaded to the IoT platform through wired and wireless networks. Basic data subscription services can be provided, and raw sensing data can be provided to the sensing agent.

[0046] Furthermore, when a perceptual agent sends its basic decision-making data to the blackboard sharing network, it can bind its agent ID (Identity document) and send it to the blackboard sharing network together with the basic decision-making data. This allows other agents to understand the specific perceptual agent through the agent ID. At the same time, when other agents need other data later, it is convenient for other agents to determine the priority notification recipients through the agent ID.

[0047] For example, taking yellow spots on rice leaves as an example, when the sensing agent detects yellow spots on rice leaves, it binds the yellow spot data to the sensing agent's agent ID (which can be automatically generated by UUID technology) and sends it to the blackboard sharing network.

[0048] Step S120: Receive the intent reasoning conclusion fed back by the intent reasoning agent based on the decision-making basis data, and publish the intent reasoning conclusion.

[0049] Intent reasoning conclusions refer to action strategies that are guided and generated based on decision-making data and the entity's own knowledge base.

[0050] Intent reasoning agents are used to infer intent reasoning conclusions based on pre-trained large models (such as plant disease and pest diagnosis models) and decision-based data, such as "disease treatment", "irrigation required", "topdressing", etc., and send the intent reasoning conclusions to the blackboard sharing network. They can also obtain the target task requirements input by other agents or users from the blackboard sharing network, thereby revising and optimizing the intent prediction.

[0051] When an intent-reasoning agent detects new decision-making data or agent IDs added to the blackboard sharing network, it reads the decision-making data from the blackboard sharing network, generates a corresponding intent-reasoning conclusion, and sends the conclusion to the blackboard sharing network. Correspondingly, the blackboard sharing network receives the intent-reasoning conclusion from the agent and publishes it.

[0052] Furthermore, when the intent reasoning agent sends the intent reasoning conclusion to the blackboard sharing network, it can bind its agent ID and send it to the blackboard sharing network together with the intent reasoning conclusion.

[0053] Furthermore, if the intent reasoning agent needs more detailed information, it can bind the required information to its own ID and send it to the blackboard sharing network. It will also prioritize notifying the perception agent by default the ID input source it reads (i.e., the agent ID bound to the decision-making base data), thus instructing the perception agent to access more refined data and higher-precision sensors for finer-grained perception. Correspondingly, the perception agent can obtain the detailed information it needs from the blackboard sharing network, perform corresponding data collection and processing, and send the data to the blackboard sharing network. Conversely, the intent reasoning agent can obtain the further detailed data uploaded by the perception agent and upload its final intent reasoning conclusion to the blackboard sharing network.

[0054] For example, in the above example, if the decision-making basis data is yellow spot data, the intention reasoning agent can determine the possible causes of yellow spot disease as: disease factors, insect factors, nutrient deficiency, pesticide damage factors, water factors, and soil factors. By obtaining detailed data and conducting multiple rounds of screening, it is determined that there are three highly probable causes for yellow spots on leaves: (1) leaf sheath rot, (2) nitrogen, magnesium, and potassium deficiency, and (3) temporary water shortage, and the intention reasoning conclusion is sent to the blackboard sharing network.

[0055] Step S130: Receive the task orchestration result fed back by the task orchestration agent based on the decision-making data and the intent reasoning conclusion.

[0056] The task orchestration agent is used to decompose tasks into operable subtasks and perform adaptive orchestration based on the system environment status, such as decision-making data and intent reasoning conclusions on the blackboard shared network.

[0057] Task orchestration results refer to the sequence of specific, executable, and time-ordered operations that the task orchestration agent transforms from decision-making data and intent reasoning conclusions into.

[0058] When the task orchestration agent detects new decision-making data, intent reasoning conclusions, and / or agent IDs in the blackboard sharing network, it obtains the decision-making data and intent reasoning conclusions from the blackboard sharing network, determines the tasks to be executed based on the intent reasoning conclusions, and then performs each task to be executed into the smallest operable granularity.

[0059] Next, the task orchestration agent uses an adaptive task orchestration algorithm to orchestrate the tasks to be executed based on the decision-making foundation data, thus obtaining the task orchestration result. Specifically, the adaptive task orchestration algorithm can calculate the execution priority score of each task to be executed, and sort the tasks to be executed based on the execution priority scores to obtain the task orchestration result.

[0060] Finally, the task orchestration agent sends the task orchestration results to the blackboard sharing network. Correspondingly, the blackboard sharing network receives the task orchestration results from the task orchestration agent and publishes them.

[0061] Furthermore, when a task orchestration agent sends the task orchestration results to the blackboard sharing network, it can bind its agent ID and send it to the blackboard sharing network along with the task orchestration results.

[0062] The information decision-making method provided in this application is applied to a blackboard sharing network. The blackboard sharing network first receives and publishes basic decision-making data sent by the perceptual agents. This means that the system's decision-making process is driven by real-world data changes rather than preset instructions, solving the dependence on predetermined tasks in existing technologies. This allows the system to adaptively discover and respond to unexpected new tasks in the environment, greatly enhancing its adaptability to complex and dynamic environments. Subsequently, the blackboard sharing network receives and publishes intent reasoning conclusions fed back by the intent reasoning agent based on the aforementioned data, and task orchestration results fed back by the task orchestration agent based on the basic decision-making data and intent reasoning conclusions. Through this approach, an asynchronous, decoupled, data-driven information decision-making method based on a blackboard sharing network is constructed. Each agent, as an independent decision-making unit, independently performs information reading and writing, data perception, reasoning, and planning around the blackboard sharing network, realizing adaptive collaboration among multiple agents. Ultimately, without human intervention, it efficiently and flexibly responds to complex and dynamic task requirements, significantly improving the flexibility and execution efficiency of complex information decision-making task orchestration.

[0063] Based on any of the above embodiments Figure 2 This is the second flowchart illustrating the information decision-making method provided in the embodiments of this application. (Refer to...) Figure 2 After step S120 or step S130, the information decision-making method may further include steps S140, S150 and S160.

[0064] Step S140: When a conflict is detected between the intention reasoning conclusion and the task orchestration result, a conflict event is issued.

[0065] When multiple agents execute in parallel, conflicts inevitably occur, mainly categorized into three typical conflict problems: control conflicts, task conflicts, and read-write conflicts. Taking agricultural information decision-making scenarios as an example, control conflicts refer to multiple agents attempting to control the same physical device to perform different operations. For instance, multiple agents may need to control a spraying drone to perform tasks on different plots. Task conflicts refer to multiple agents controlling different devices to perform conflicting operations on the same crop. For example, agent A may need to increase the amount of fertilizer, while agent B may need to decrease it. Read-write conflicts refer to multiple agents simultaneously modifying content written by a single agent. For example, an intent-reasoning agent publishes a case of fertilizer deficiency on a shared network, while other agents, based on their own reasoning, arrive at different intent-reasoning conclusions and attempt to simultaneously notify the same agent to change the diagnosis to pesticide or water deficiency. To address these agricultural information decision-making conflict scenarios, this application's embodiments design a weighted time-locking method combined with a decision-arbitration agent to resolve the conflict problem.

[0066] When the blackboard sharing network detects a conflict between the intent reasoning conclusion and the task orchestration result, it publishes the conflict event to all agents and marks the conflict content or request as "pending arbitration". At the same time, the conflict content or request can be transferred to the cache area (cache area) along with the conflict type and conflict information, which includes, but is not limited to, the agent IDs of the two conflicting parties and the time when the conflict event occurred.

[0067] Step S150: Receive the arbitration result corresponding to the conflict event sent by the decision arbitration agent. The arbitration result includes the winning agent and the lock time.

[0068] The decision-arbitrary agent is used to negotiate and collaborate on tasks, prioritize tasks, and arbitrate conflicts when multiple agents conflict over modifications to shared information or when they need to operate on the same physical facilities. The agent with higher priority than other agents publishes its decision to the blackboard sharing network, allowing other agents to adjust their strategies accordingly.

[0069] Specifically, the decision-making arbitration agent monitors the content on the blackboard sharing network. When a conflict event is detected, it determines the corresponding arbitration result based on the conflict event. The arbitration result includes the winning agent and the lock time. Then, the arbitration result is sent to the blackboard sharing network. Correspondingly, the blackboard sharing network receives the arbitration result corresponding to the conflict event sent by the decision-making arbitration agent.

[0070] Step S160: Based on the arbitration result, issue a lock announcement. The lock announcement is used to indicate that the content of the winning agent is valid during the lock time and to instruct the unsuccessful agents to enter a backoff waiting state.

[0071] The blackboard sharing network issues a lock announcement based on the arbitration result. The lock announcement indicates that the content of the winning agent is valid for the lock period and instructs the unsuccessful agents to enter a backoff waiting state. That is, during the lock period, only the content (intent reasoning conclusion or task orchestration result) published by the winning agent is valid. Unsuccessful agents, upon detecting the lock announcement, enter a backoff waiting state and will not resubmit content during the lock period.

[0072] The information decision-making method provided in this application introduces a decision arbitration agent and a conflict arbitration mechanism, which solves the problems of information synchronization and conflict management in multi-agent collaboration and greatly enhances the robustness of the system.

[0073] Based on any of the above embodiments, the arbitration result is determined through the following steps S11, S12 and S13.

[0074] Step S11: Based on the conflict event, obtain the target conflict agent and its current weight, the rationality score of the conflict content, and the time difference between the occurrence time of the conflict event and the current time.

[0075] The decision-making arbitration agent identifies the target conflict agent based on conflict events, and simultaneously obtains the target conflict agent's current weight, the rationality score of the conflict content, and the time difference between the occurrence time of the conflict event and the current time. The current weight can be a fixed preset value or dynamically adjusted through learning; the rationality score can be determined by prediction accuracy, human feedback, etc.

[0076] Step S12: Calculate the arbitration priority score for each of the target conflict agents based on the current weight, the rationality score, and the time difference.

[0077] Then, the decision-making arbitration agent calculates the priority score of each target conflict agent based on the current weight, rationality score and time difference, which is recorded as the arbitration priority score.

[0078] Assume that the collision occurs in the blackboard sharing network as follows: Among them, the conflicting term C i (1≤i≤n) corresponds to the The calculation formula for the arbitration priority score for each target conflicting agent is as follows: ; in, express Arbitration priority scoring at time t Indicates the current weight. Indicates the reasonableness score. Indicates time difference, This represents the time decay factor, used to control the priority of new tasks.

[0079] Step S13: Select the agent with the highest arbitration priority score as the winning agent, and calculate the locking time based on the arbitration priority score of the winning agent.

[0080] Content with higher arbitration priority scores should be adopted first. Therefore, the agent with the highest arbitration priority score is selected as the winning agent, and the lock-in time is calculated based on the winning agent's arbitration priority score. The formula for calculating the lock-in time is as follows: ; in, Indicates the lock time of the winning agent; and These represent the minimum lock time and the maximum lock time, respectively, and can be dynamically adjusted based on historical records. The score represents the arbitration priority score of the winning agent; θ represents the adjustment parameter used to control the inverse relationship between the arbitration priority score and the lock time. A higher score corresponds to a shorter lock time, which is convenient for iterative improvement. A lower score requires waiting for a longer time before trying to re-propose.

[0081] The information decision-making method provided in this application offers a quantitative, interpretable, and multi-dimensional arbitration priority scoring algorithm. It comprehensively considers the agent's historical credibility (i.e., current weight), the rationality of the current suggestion (i.e., rationality score), and the timeliness of the response (i.e., time difference), ensuring that arbitration decisions are rational judgments based on current data and historical performance, significantly improving the scientific rigor and accuracy of the decision. Furthermore, this application uses the arbitration priority scoring result as the core, while dynamically adjusting the locking time based on historical records, achieving system load balancing and conflict minimization.

[0082] Based on any of the above embodiments, after step S160, the information decision-making method further includes steps S170 and S180.

[0083] Step S170: Obtain the content feedback score of the target conflict agent and obtain the average feedback score.

[0084] Content feedback score refers to the feedback score for conflicting content. Specifically, it can be a human feedback score or a task result feedback score. It can be obtained by the device-controlled intelligent agent after performing the task corresponding to the conflicting event, or it can be obtained by the feedback learning intelligent agent after the task corresponding to the conflicting event is completed, by comparing the basic decision data and the corresponding completion status data (i.e., effect verification data).

[0085] The average feedback score is the average of the feedback scores of all agents.

[0086] Step S180: Receive the updated weights sent by the feedback learning agent based on the content feedback score, the average feedback score, and the current weights.

[0087] The feedback learning agent updates the current weights based on content feedback scores and average feedback scores, and then sends the updated weights to the blackboard sharing network. Correspondingly, the blackboard sharing network receives and publishes the updated weights for the decision-making arbitration agent to read and use subsequently.

[0088] The updated weights are determined through the following steps S21 and S22: Step S21: Calculate the difference between the content feedback score and the average feedback score.

[0089] Step S22: Using a reinforcement learning algorithm, update the current weight of the winning agent based on the difference.

[0090] Specifically, after obtaining the content feedback score and the average feedback score, the feedback learning agent calculates the difference between the content feedback score and the average feedback score. Then, using a reinforcement learning algorithm, it updates the current weights of the winning agent based on this difference. The specific calculation formula is as follows: ; in, Indicates the updated weights; This represents the weight at time t, i.e., the current weight; Indicates the learning rate. This indicates the content feedback rating. This represents the average feedback score.

[0091] The information decision-making method provided in this application introduces a dynamic weight update mechanism, realizes a weight self-learning process based on execution effect, supports dynamic optimization of decision arbitration, and continuously optimizes task allocation efficiency.

[0092] Based on any of the above embodiments, the arbitration result also includes the backoff time of the unsuccessful agent, which is calculated based on the lock time, dynamic delay time, and random perturbation value, and the dynamic delay time is determined based on the number of historical failures of the unsuccessful agent.

[0093] Furthermore, after step S160, the method further includes steps S190 and S200.

[0094] Step S190: Receive the resubmission request triggered by the unsuccessful agent after the backoff time has elapsed.

[0095] The arbitration result includes the winning agent and the lock-in time, as well as the backoff time for the unsuccessful agents. The backoff time is the interval between when an unsuccessful agent enters a backoff waiting state and resubmits its request. The backoff time is calculated based on the lock-in time, dynamic delay time, and random perturbation value, using the following formula: ; in, L represents the backoff time; L represents the lock-in time of the winning agent. The dynamic delay time during backoff can be determined based on the number of historical failures; This represents a random perturbation value, which is relatively small and is used to break the homogeneity of decisions and prevent deadlock.

[0096] After an agent that fails to win a bid enters a backoff waiting state, it can trigger a resubmission request after the backoff time has elapsed, thus sending the proposal to the blackboard sharing network again. Correspondingly, the blackboard sharing network receives the resubmission request sent by the unsuccessful agent.

[0097] It should be understood that after entering the backoff waiting state, unsuccessful agents can continuously monitor changes in the content of the blackboard sharing network and can revise their proposals (content or requests) based on the changes before resubmitting them.

[0098] Step S200: When it is detected that there is no conflict between the resubmission requests, perform the corresponding operation based on the resubmission requests.

[0099] When the blackboard sharing network receives a resubmission request, it checks whether there are still conflicts between the resubmission requests. If no conflicts are detected, it performs the corresponding operation based on the resubmission request, such as a write operation or a task execution operation.

[0100] The information decision-making method provided in this application, through a backoff time determination method with punitive dynamic delay and random perturbation, effectively avoids the waste of system resources and deadlock conflicts, and maintains the overall stability and orderliness of the system.

[0101] Based on any of the above embodiments, before step S130, step S210 is further included.

[0102] Step S210: Receive available device information sent by the device control agent and publish the available device information.

[0103] The device control agent is used to remotely connect to the device, obtain available device information, and send it to the blackboard sharing network; it is also used to control the device to perform corresponding task operations. In agricultural information decision-making scenarios, the device refers to agricultural implements and facilities that can change the state of the crop growth environment, such as spraying and fertilizing drones, irrigation systems, lighting systems, and ventilation systems. Available device information may include, but is not limited to, the function of the implement and rule parameter information.

[0104] The device control agent monitors available devices in real time and sends this information to the blackboard sharing network as a reference for other agents' reasoning, while also constraining agents to use existing agricultural implements to perform operations. Correspondingly, the blackboard sharing network receives the available device information sent by the device control agent and publishes this information.

[0105] At this time, step S130 includes: step S131.

[0106] Step S131: Receive the task orchestration result fed back by the task orchestration agent based on the decision-making data, the intent reasoning conclusion, and the available device information.

[0107] When the task orchestration agent detects new decision-making data, intent reasoning conclusions, available device information, and / or agent IDs in the blackboard sharing network, it retrieves these data from the blackboard sharing network to determine the task orchestration result. Then, it sends the task orchestration result to the blackboard sharing network. Correspondingly, the blackboard sharing network receives the task orchestration result from the task orchestration agent and publishes it.

[0108] Furthermore, the task orchestration result is obtained by the task orchestration agent through the following steps S31, S32 and S33.

[0109] Step S31: Based on the intent reasoning conclusion, determine the task to be executed, and decompose the task to be executed to obtain sub-tasks to be executed.

[0110] First, based on the pre-defined mapping relationship between the intention reasoning conclusion and the task, the task corresponding to the intention reasoning conclusion is determined and recorded as the task to be executed.

[0111] Then, the task to be executed is decomposed, that is, it is broken down into the smallest operable granularity to obtain the subtasks to be executed.

[0112] For example, for the three intentional inference conclusions in the above example: (1) leaf sheath rot, (2) nitrogen, magnesium and potassium deficiency, (3) temporary water shortage, the tasks to be performed can be determined as spraying, fertilizing and irrigating, respectively. The results after decomposing the tasks to be performed are as follows: (1) Leaf sheath rot: ①Activate the sprinkler system; ②Inject water; ③Inject tebuconazole; ④Stir and mix; ⑤Activate the atomizer; ⑥Transmit the plot coordinates; ⑦Execute sprinkler irrigation; (2) Nitrogen, magnesium, and potassium deficiency: ① Activate the sprinkler irrigation system; ② Inject water; ③ Inject a mixture of nitrogen, magnesium, and potassium or a mixture of all three nutrients; ④ Stir and mix; ⑤ Activate the drip irrigation system; ⑥ Transmit the plot coordinates; ⑦ Execute sprinkler irrigation; (3) Temporary water shortage: ① Call the sprinkler system; ② Inject water; ③ Call the drip irrigation system; ④ Transmit the plot coordinates; ⑤ Execute sprinkler irrigation.

[0113] Step S32: Based on the decision-making data and the available device information, calculate the execution priority score of the task to be executed.

[0114] Then, based on the decision-making data and available equipment information, the priority score of each task to be executed is calculated and recorded as the execution priority score.

[0115] When calculating the execution priority score, factors such as task urgency, land parcel status, historical policy implementation, and available resources can be considered to perceive the land parcel status in real time and provide feedback for adaptive scheduling adjustments. The specific algorithm is as follows: Generate a list of tasks to be executed, T={T1,T2,...,T...} n Then, based on the decision-making data and available equipment information, the urgency, scope of impact, estimated execution time, and importance of each task to be executed are determined. These are then weighted and summed to obtain an execution priority score, as shown in the following formula: ; Among them, P i E represents the execution priority score of the i-th task to be executed. i A represents the urgency level of the i-th task to be executed. i D represents the scope of influence of the i-th task to be executed. i S represents the estimated execution time of the i-th task to be executed. i The importance of the i-th plot to be executed is represented by α, β, μ and ν, which represent weight coefficients. The weight coefficients can be obtained through expert experience in the early stage of task scheduling and can be continuously optimized through feedback and learning in the later stage.

[0116] Of course, it is understandable that specific scoring rules can be set according to actual needs and are not limited to the rules mentioned above.

[0117] Step S33: Based on the execution priority score, the available device information, and the sub-tasks to be executed, sort and allocate resources for the tasks to be executed to obtain the task orchestration result.

[0118] Based on execution priority scores, available device information, and pending subtasks, the tasks to be executed are sorted and resources are allocated to obtain the task orchestration result. That is, according to P... i The task flowchart is executed sequentially according to the order of tasks to be executed. If the available device information indicates that resources allow, parallel tasks are supported.

[0119] For example, let α=0.4, β=0.3, μ=0.2, ν=0.1, and let leaf sheath rot, nitrogen, magnesium, and potassium deficiency, and temporary water shortage correspond to tasks T1, T2, and T3 respectively. The values ​​for task T1 are E1=0.9, A1=0.6, D1=15, and S1=0.7. Then P1=0.4×0.9+0.3×0.6+0.2×(1 / 15)+0.1×0.7≈0.774; the values ​​for task T2 are... Given E2=0.5, A2=0.8, D2=12, and S2=0.6, then P2=0.4×0.5+0.3×0.8+0.2×(1 / 12)+0.1×0.6≈0.604; for task T3 to be executed, given E3=0.4, A3=0.5, D3=5, and S3=0.5, then P3=0.4×0.4+0.3×0.5+0.2×(1 / 5)+0.1×0.5≈0.496. At this point, the task order is: T1>T2>T3. Then, resource binding is performed on the subtasks to be executed based on available device information. The resulting task orchestration is a specific, executable, time-sequential, and resource-bound sequence of operations.

[0120] The information decision-making method provided in this application involves a task orchestration agent dynamically detecting available device information and combining it with decision-making baseline data to calculate execution priority scores, thereby ranking the tasks to be executed. Simultaneously, based on available device information, each sub-task to be executed is bound to an execution task, ultimately yielding the task orchestration result. This method effectively links abstract intent reasoning conclusions with specific device resources. By dynamically acquiring available device information, task orchestration is no longer merely theoretically feasible, but ensures that each generated task is physically feasible. Furthermore, the comprehensive ranking based on priority and resource status enables optimized scheduling of limited device resources, avoiding resource conflicts and improving overall operational efficiency.

[0121] Based on any of the above embodiments, after step S130 or step S131, the information decision-making method further includes: steps S220, S230 and S240.

[0122] Step S220: Receive the task execution result and effect verification request fed back by the device control agent based on the task orchestration result, and publish the task execution result and the effect verification request.

[0123] The device control agent is also used to obtain task orchestration results from the blackboard sharing network, control relevant devices to perform corresponding operations based on the task orchestration results, and then send the execution results to the blackboard sharing network. Furthermore, if it is necessary to understand the effect after execution, the device control agent can send an effect verification request to the blackboard sharing network. When sending the effect verification request, it can bind its own agent ID, and simultaneously bind the agent ID of the sensing agent.

[0124] Correspondingly, the blackboard sharing network receives the task execution results and effect verification requests from the device control agent based on the task orchestration results, and then publishes the task execution results and effect verification requests.

[0125] For example, taking the above case of yellow spots on leaves as an example, the farm tools are called to perform operations according to the priority of tasks to be executed T1>T2>T3. To reduce resource contention and optimize resource locks, operations with the same properties and no relation can be merged, operations with the same properties and related properties are executed sequentially according to priority, and different operations can be executed in parallel. When resources are sufficient, multiple sets of operations can be abstracted, and all operations can be executed in parallel. Taking the smallest set of operations as an example, a typical sequence diagram is as follows: Figure 3 As shown, the device control agent controls agricultural implements to complete tasks on the corresponding plot of land and shares the execution status on the blackboard sharing network. Simultaneously, it binds its own ID and writes new requirements, such as monitoring the leaf condition of the corresponding plot hourly. The sensing agent obtains this new requirement from the blackboard sharing network, continues to sense the leaf condition, and publishes it to the blackboard sharing network.

[0126] Step S230: Receive effect verification data sent by the perceptual agent based on the effect verification request, and publish the effect verification data.

[0127] After receiving an effect verification request, the perceptual agent analyzes the data, collects relevant information, records it as effect verification data, and sends it to the blackboard sharing network. Correspondingly, the blackboard sharing network receives the effect verification data sent by the perceptual agent based on the effect verification request and publishes the effect verification data.

[0128] Step S240: Receive model parameters from the feedback learning agent based on the effect verification data, the decision-making basis data, and the task execution results. The model parameters are used to calculate the arbitration priority score.

[0129] The feedback learning agent is used for post-task monitoring, acquiring information on state changes before and after task execution (i.e., basic decision data and effect verification data). It then uses attribution analysis and reinforcement learning mechanisms to evaluate operational effectiveness. If the effectiveness is unsatisfactory, the system can automatically revert, adjust recommendations, or fine-tune model parameters, sharing this information on the blackboard sharing network for other agents to optimize strategies and adjust parameters for subsequent executions. The feedback learning agent subscribes to all data related to the task process. After acquiring effect verification data, basic decision data, and task execution results, it analyzes this data to obtain updated model parameters, which are then fed back to the blackboard sharing network. Correspondingly, the blackboard sharing network receives and publishes the model parameters fed back by the feedback learning agent; these parameters are used to calculate the arbitration priority score.

[0130] In the example above, the feedback learning agent obtains information about leaf changes from the blackboard sharing network. Based on comparisons across different plots, it can quickly and effectively determine the cause of yellow spot disease on leaves (assuming it's a deficiency of nitrogen, magnesium, and potassium). This information is then communicated to the blackboard sharing network, and the agricultural implement control agent applies fertilizer to all plots according to task T2 (nitrogen, magnesium, and potassium deficiency). It should be understood that actual situations are more complex, potentially involving the lack of a single nutrient or multiple nutrients with varying concentration requirements. Therefore, multiple small plots can be added for comparison during task execution, with multiple rounds of feedback. The feedback learning agent can continuously optimize its understanding of the perceived content and its control over the agricultural implements through reinforcement learning. It stores various situations, facilitating the reuse of agricultural implement control parameters for similar problems, reducing the overall computational load of the system, and continuously optimizing through feedback learning to form a more accurate and effective multi-agent collaborative system.

[0131] Specifically, the reinforcement learning model of the feedback learning agent can be modeled as follows: ; Where M represents the model, S represents the state space, A represents the action space, T represents the state transition function, R represents the reward function, and γ represents the reward.

[0132] The goal of the above modeling is to enable the agent to learn a policy. When faced with crop anomalies such as yellow spots, it can optimally select actions. The state space S represents each plot i, time t, and state includes leaf state features recognized from image analysis (such as the area of ​​yellow spots). Soil nutrient vectors (including nitrogen, magnesium, and potassium concentrations) Previous fertilization record Perceive historical labels (such as whether the intervention in the front wheel was effective). .Right now: Action space A can be single or compound fertilization: , where c NThis indicates the amount of nitrogen fertilizer applied, c M This indicates the amount of phosphate fertilizer applied, c K This represents the amount of potassium fertilizer applied. The state transition function T can be estimated through control plot experiments and the perception mechanism of the sensing agent, for a set of plots with similar symptoms. Perform different actions By comparing the state change results received by the feedback learning agent, the expected impact of different actions on crop health changes is estimated. The reward function R aims to maximize crop recovery after intervention and minimize fertilization costs. Let the leaf health score of the plot (generated by visual recognition) be: The current fertilization cost is [value], and the degree of health improvement after intervention is [value]. The reward function is then... Fertilization costs can be set as follows: Substituting into the reward function, we get: Among the parameters The reward γ can be initialized with expert experience in the initial stage and automatically trained and adaptively adjusted during system operation. In each round of feedback, the feedback learning agent responds to different actions. Record the reward on plot j. Select the action with the highest expected value for broadcasting to other similar plots: ; Simultaneously, the action execution and state update are recorded in the blackboard sharing network. Policy updates can be performed using reinforcement learning policy functions such as Q-Learning and DQN (Deep Q-Network).

[0133] For the case of yellow spots on leaves, the procedure was as follows: (1) The sensing agent reads visual and soil data and constructs state S; (2) The feedback learning agent reads the macular abnormality status from the blackboard sharing network; (3) Based on The strategy involves selecting multiple actions to test on small plots (actions) ); (4) Compare the feedback (health improvement) of each plot, according to Estimate the reward; (5) Optimal action The information is written into the blackboard sharing network, and the fertilization equipment is controlled by the device control intelligent agent. (6) The feedback result serves as the status for the next round of feedback. Used for the next update.

[0134] The information decision-making method provided in this application constructs a complete and advanced intelligent closed loop from execution to evaluation to optimization. The system can not only complete tasks, but also learn and reflect from the task results, update model parameters, and dynamically optimize its future behavior strategies, thereby solving the problem of lack of feedback optimization in existing systems.

[0135] Figure 4 This is the third flowchart illustrating the information decision-making method provided in the embodiments of this application. (Refer to...) Figure 4 This application provides an information decision-making method, which is applied to an information decision-making system. The information decision-making system includes a blackboard sharing network, a perception agent, an intent reasoning agent, and a task orchestration agent. The information decision-making method includes steps S310, S320, and S330.

[0136] In step S310, the blackboard sharing network receives the decision-making basis data sent by the perceptual agent.

[0137] In this embodiment, the information decision-making method is applied to an information decision-making system, which includes a blackboard sharing network, a perceptual agent, an intent reasoning agent, and a task orchestration agent.

[0138] The perceptual agent sends the basic decision data to the blackboard sharing network. Correspondingly, the blackboard sharing network receives the basic decision data sent by the perceptual agent and publishes it. The publishing methods include, but are not limited to: (1) writing it into the blackboard data area and making it public to all agents; (2) writing it into the blackboard data area and making it public to agents that have subscribed to the basic decision data; (3) writing it into the blackboard data area and sending it to agents that have subscribed to the basic decision data.

[0139] Furthermore, when a perceptual agent sends decision-making data to the blackboard sharing network, it can bind its agent ID and send it to the blackboard sharing network along with the decision-making data. This allows other agents to understand the specific perceptual agent through the agent ID, and when other agents need other data later, it is convenient for them to determine the priority notification recipients through the agent ID.

[0140] In step S320, the intention reasoning agent obtains the decision-making basis data from the blackboard sharing network, feeds back the intention reasoning conclusion based on the decision-making basis data, and feeds back the intention reasoning conclusion to the blackboard sharing network.

[0141] When an intent-reasoning agent detects new decision-making data or agent IDs added to the blackboard sharing network, it reads the decision-making data from the blackboard sharing network, generates a corresponding intent-reasoning conclusion, and sends the conclusion to the blackboard sharing network. Correspondingly, the blackboard sharing network receives the intent-reasoning conclusion from the agent and publishes it.

[0142] In step S330, the task orchestration agent obtains the decision-making basis data and the intent reasoning conclusion from the blackboard sharing network, determines the task orchestration result based on the decision-making basis data and the intent reasoning conclusion, and feeds back the task orchestration result to the blackboard sharing network.

[0143] When the task orchestration agent detects new decision-making data, intent reasoning conclusions, and / or agent IDs in the blackboard sharing network, it retrieves the decision-making data and intent reasoning conclusions from the blackboard sharing network. Based on the intent reasoning conclusions, it determines the tasks to be executed and then breaks down each task into its smallest operable granularity. Next, the task orchestration agent uses an adaptive task orchestration algorithm to orchestrate the tasks based on the decision-making data, obtaining the task orchestration result. Specifically, the adaptive task orchestration algorithm calculates the execution priority score of each task, sorts the tasks based on the execution priority score, and obtains the task orchestration result. Finally, the task orchestration agent sends the task orchestration result to the blackboard sharing network. Correspondingly, the blackboard sharing network receives the task orchestration result from the task orchestration agent and publishes the task orchestration result.

[0144] The information decision-making method provided in this application is applied to an information decision-making system, which includes a blackboard sharing network, a perceptual agent, an intent reasoning agent, and a task orchestration agent. The blackboard sharing network first receives and publishes basic decision-making data sent by the perceptual agent. This means that the system's decision-making process is driven by real-world data changes rather than preset instructions, solving the dependence on predetermined tasks in existing technologies. This allows the system to adaptively discover and respond to unexpected new tasks in the environment, greatly enhancing its adaptability to complex and dynamic environments. Subsequently, the intent reasoning agent obtains basic decision-making data from the blackboard sharing network to determine the intent reasoning conclusion and feeds it back to the blackboard sharing network. Finally, the task orchestration agent obtains the basic decision-making data and the intent reasoning conclusion from the blackboard sharing network, determines the task orchestration result based on the basic decision-making data and the intent reasoning conclusion, and feeds the task orchestration result back to the blackboard sharing network. Through the above approach, an asynchronous, decoupled, data-driven information decision-making method based on a blackboard sharing network was constructed. Each agent, as an independent decision-making unit, independently performs information reading and writing, data perception, reasoning, and planning around the blackboard sharing network, realizing adaptive collaboration among multiple agents. Ultimately, without human intervention, it can efficiently and flexibly cope with complex and dynamic task requirements, significantly improving the flexibility and execution efficiency of complex information decision-making task orchestration.

[0145] Figure 5 This is a schematic diagram of the structure of the information decision-making system provided in an embodiment of this application. (Refer to...) Figure 5 This application provides an information decision-making system. The information decision-making system provided in this application is described below. The information decision-making system described below can be referred to in correspondence with the information decision-making method described above.

[0146] The information decision-making system includes: a blackboard sharing network, a perception agent, an intent reasoning agent, and a task orchestration agent.

[0147] The blackboard sharing network is used to execute the aforementioned information decision-making method.

[0148] The sensing agent is used to collect basic decision-making data and send the basic decision-making data to the blackboard sharing network.

[0149] The intent reasoning agent is used to obtain the decision-making basis data from the blackboard sharing network, determine the intent reasoning conclusion based on the decision-making basis data, and feed back the intent reasoning conclusion to the blackboard sharing network.

[0150] The task orchestration agent is used to obtain the decision-making basis data and the intent reasoning conclusion from the blackboard sharing network, determine the task orchestration result based on the decision-making basis data and the intent reasoning conclusion, and feed back the task orchestration result to the blackboard sharing network.

[0151] Traditional information decision-making systems typically employ a centralized control architecture. Taking a traditional agricultural information decision-making system as an example, firstly, data on crops and the environment, such as temperature, humidity, soil conditions, and images of pests and diseases, are collected through manual inspections or simple sensor devices. This data is then uploaded to a central server where agricultural experts or technicians analyze and assess it to identify problems such as pests and diseases, insufficient water, or nutrient deficiencies. Experts then develop corresponding treatment plans, such as arranging irrigation, fertilization, or pesticide spraying. Finally, farmers manually operate agricultural equipment to complete the tasks according to the plan. After the tasks are completed, the effectiveness is usually evaluated through manual observation or resampling. This model is slow to respond, relies on human experience, lacks intelligent reasoning and automation capabilities, has low coordination between devices, and has poor system scalability and fault tolerance.

[0152] With the development of large-scale AI modeling technology, flowchart-based workflow orchestration is currently being implemented. Tasks are graphically linked using pre-set workflow templates, and the system automatically executes the workflow based on trigger events. When a sensor detects an anomaly (such as excessively high temperature or yellowing leaves), the system matches the corresponding workflow template, such as "disease detection → diagnosis → spraying," and sequentially schedules relevant equipment to perform the task. The entire process is highly automated and suitable for standardized, repetitive agricultural scenarios, but it struggles to handle unexpected situations or complex environmental changes. The system lacks intelligent reasoning capabilities and cannot dynamically generate new tasks based on real-time data.

[0153] Therefore, this application provides a more flexible and advanced workflow for agricultural information decision-making systems, constructing a decentralized information decision-making system composed of multiple agents. The entire system does not rely on fixed process templates but is based on an event-driven mechanism. Agents autonomously determine whether to intervene in task processing based on changes in information on the shared network blackboard. This mechanism possesses high flexibility and adaptability, enabling dynamic responses to complex and ever-changing agricultural environments. Simultaneously, the system supports reinforcement learning and has continuous optimization capabilities. Due to the loose coupling between agents, the system is highly fault-tolerant; the failure of any single agent will not affect the overall operation.

[0154] Specifically, the information decision-making system includes a blackboard sharing network, a perceptual agent, an intent reasoning agent, and a task orchestration agent. The blackboard sharing network executes the aforementioned information decision-making method. The perceptual agent collects basic decision-making data and sends it to the blackboard sharing network. The intent reasoning agent obtains the basic decision-making data from the blackboard sharing network, determines the intent reasoning conclusion based on the data, and feeds the conclusion back to the blackboard sharing network. The task orchestration agent obtains both the basic decision-making data and the intent reasoning conclusion from the blackboard sharing network, determines the task orchestration result based on these data, and feeds the result back to the blackboard sharing network. The specific execution process can be found in the above embodiments and will not be elaborated here.

[0155] The information decision-making system provided in this application is a multi-agent information decision-making system with adaptive task orchestration capabilities. The system consists of multiple functionally independent intelligent agents, each undertaking functions such as perception, analysis, reasoning, and task orchestration. It adopts a distributed and decentralized design, with each agent possessing independent operating capabilities and not relying on central instructions. This results in higher system stability and scalability; even if individual modules fail, the overall system can still maintain operation. Compared to traditional information decision-making systems that rely on preset templates or fixed processes, the information decision-making system in this application has adaptive task orchestration capabilities. It can dynamically generate optimal task orchestration results based on multi-dimensional factors such as basic decision-making data (e.g., crop status, agricultural resources, real-time environmental changes) and intent reasoning conclusions. Facing common uncertainties in agriculture (e.g., sudden weather changes, pest and disease outbreaks, market price fluctuations), the system can autonomously judge and adjust task strategies, achieving flexible response and accurate decision-making. Furthermore, the information decision-making system in this application is based on a modular design; new intelligent agents or functional modules can be easily connected to the blackboard sharing network to expand the system's functionality. The system can flexibly adjust the functions of intelligent agents and task orchestration strategies according to different scenarios and needs.

[0156] Furthermore, the information decision-making system also includes a decision arbitration intelligent agent.

[0157] The blackboard sharing network is used to publish a conflict event when a conflict is detected between the intention reasoning conclusion and the task orchestration result.

[0158] The decision-making arbitration agent is used to determine the arbitration result corresponding to the conflict event and send the arbitration result to the blackboard sharing network; wherein, the arbitration result includes the winning agent and the lock time.

[0159] The blackboard sharing network is also used to issue a lock announcement based on the arbitration result. The lock announcement is used to indicate that the content of the winning agent is valid during the lock time and to instruct the unsuccessful agents to enter a backoff waiting state.

[0160] The information decision-making system provided in this application embodiment further includes a decision-arbitration agent. When the blackboard sharing network detects a conflict between the intent reasoning conclusion and the task orchestration result, it publishes a conflict event. The decision-arbitration agent determines the arbitration result corresponding to the conflict event and sends the arbitration result to the blackboard sharing network. The arbitration result includes the winning agent and the lock time. Based on the arbitration result, the blackboard sharing network publishes a lock announcement. The lock announcement indicates that the content of the winning agent is valid within the lock time and instructs the unwinning agents to enter a backoff waiting state. By introducing a decision-arbitration agent into the system, conflicts in tasks are coordinated, ensuring the stable and efficient operation of the multi-agent information decision-making system in complex agricultural scenarios. The system optimizes resource allocation through dynamic task orchestration and conflict arbitration mechanisms, avoiding resource contention and waste, and improving the efficiency of task operations.

[0161] Furthermore, the information decision-making system also includes a device control agent and a feedback learning agent.

[0162] The device control agent is used to obtain the task orchestration results from the blackboard sharing network, execute the task, and send the task execution results and effect verification requests to the blackboard sharing network.

[0163] The sensing agent is also used to collect effect verification data and send the effect verification data to the blackboard sharing network when it receives the effect verification request from the blackboard sharing network.

[0164] The feedback learning agent is used to obtain the decision-making basis data, the task execution results, and the effect verification data from the blackboard sharing network, and to optimize the model parameters using reinforcement learning algorithms.

[0165] The information decision-making system provided in this application embodiment further includes a device control agent and a feedback learning agent. The device control agent executes tasks based on task encoding results, while the feedback learning agent implements a mechanism for verifying the effect after task completion, building a closed-loop adaptive evolution capability that can continuously learn and improve system performance. Specifically, after the device control agent executes a task, the sensing agent collects feedback information (such as changes in crop state, environmental responses, etc.) and continuously optimizes the task strategy through the reinforcement learning mechanism of the feedback learning agent, achieving learning on the job. Traditional systems end once the task is completed, without knowing whether the decision is effective. However, the information decision-making system provided in this application embodiment observes the results after each decision, judges whether it is successful, and performs better when encountering similar situations in the future. This helps to accumulate knowledge, reduces reliance on human experience, realizes the self-evolution of the information decision-making system, optimizes device control parameters, and reduces the overall computational load of the system.

[0166] Furthermore, the blackboard sharing network can include a blackboard area, a cache area, monitors, publishers, and controllers. The blackboard area stores the shared state and information of various agents. Each sub-block corresponds to a specific agent or task domain, recording its relevant data. All data carries a unique agent ID, timestamp, and version information to support collaborative access and version management. The cache area temporarily stores data that may have write conflicts, is unconfirmed, or awaits arbitration. It can also serve as an intermediate result storage area for inference of candidate solutions and preprocessing analysis. Cache data is uniformly labeled with conflict markers and the source agent to facilitate arbitration decisions. Monitors are responsible for real-time monitoring of registration, write, and update requests from all agents. Upon detecting data changes, they can automatically drive the publisher or controller to respond. The publisher is responsible for pushing data change events and supports fine-grained subscription mechanisms based on topics, regions, and data types. For example, an agent can subscribe only to "crop disease diagnosis results" or "job queues for a specified plot." The publisher can also synchronize conflict information from the cache area to all subscribers for collaborative response. The controller is responsible for managing blackboard read / write permissions, agent registration and ID allocation, data consistency maintenance, conflict detection and arbitration execution, and also records all operation history for easy backtracking.

[0167] Taking the appearance of yellow spots on crop leaves in an agricultural information decision-making scenario as an example, the collaborative process of each module in the blackboard sharing network is illustrated as follows: Phase 1: Initialization and Mounting.

[0168] (1) The blackboard sharing network is initialized, and various intelligent agents (such as sensing intelligent agents, decision-making and arbitration intelligent agents, and device control intelligent agents) are started.

[0169] (2) The controller completes the access authentication of the intelligent agent and assigns a unique ID to each intelligent agent.

[0170] (3) The monitor receives the blackboard mounting request from the intelligent agent and forwards it to the publisher.

[0171] (4) The publisher will write the mounted metadata into the blackboard block, such as agent ID, responsibility, subscribed topics, etc.

[0172] (5) Blackboard state change → Monitor senses → Publisher notifies all agents of blackboard update.

[0173] (6) Each intelligent agent completes its attachment and forms a preliminary cooperative network.

[0174] Phase Two: Perception and Anomaly Detection.

[0175] (7) The sensing agent periodically collects and updates the state of crop leaves and continuously writes them into the blackboard data area.

[0176] (8) When an anomaly is detected (such as yellow spots on the leaves), the data change is detected by the blackboard monitor.

[0177] (9) Related intelligent agents (such as intention reasoning intelligent agents, task orchestration intelligent agents, feedback learning intelligent agents, and device control intelligent agents) learn about events through the subscription mechanism and enter a collaborative state.

[0178] Phase Three: Task Coordination and Conflict Arbitration.

[0179] (10) Multiple agents need to call the same type of agricultural implements (such as irrigation systems) and issue control requests in their respective blocks.

[0180] (11) The monitor detected a write target conflict (i.e. multiple agents compete for the same irrigation equipment).

[0181] (12) Conflict requests are moved to the cache area, along with the conflict type and information about the parties involved.

[0182] (13) The publisher notifies all agents of the conflict event and marks it as "pending arbitration".

[0183] Phase Four: Arbitration and Dispatch Execution.

[0184] (14) The decision-making arbitration agent makes an arbitration plan based on factors such as task urgency, land priority, and agent weight (such as lock time and waiting execution order).

[0185] (15) The arbitration result is broadcast by the publisher.

[0186] (16) After receiving the result, the conflicting agent adjusts the control request and resubmits it (e.g., postpones execution).

[0187] (17) Until a certain moment there is only one valid request, the agricultural implement control agent acquires control of the equipment and begins to perform the irrigation task.

[0188] After the irrigation task is completed, the equipment control agent issues a lock timeout notification on the blackboard, and subsequent conflict requests will be reinstated.

[0189] The information decision-making apparatus provided in the embodiments of this application is described below. The information decision-making apparatus described below and the information decision-making method described above can be referred to and correspond to each other.

[0190] Figure 6 This is a schematic diagram of the structure of the information decision-making device according to an embodiment of this application, with reference to... Figure 6 This application embodiment also provides an information decision-making device, which may include: a basic data receiving module 610, a reasoning conclusion receiving module 620, and an arrangement result receiving module 630; wherein: The basic data receiving module 610 is used to receive the basic decision data sent by the perceptual intelligent agent and to publish the basic decision data. The reasoning conclusion receiving module 620 is used to receive the intention reasoning conclusion fed back by the intention reasoning agent based on the decision-making basis data, and to publish the intention reasoning conclusion; The orchestration result receiving module 630 is used to receive the task orchestration result fed back by the task orchestration agent based on the decision-making basis data and the intent reasoning conclusion.

[0191] The information decision-making device provided in this application embodiment can be a blackboard sharing network. First, it receives and publishes basic decision-making data sent by the perceptual agent. This means that the system's decision-making process is driven by real-world data changes rather than preset instructions, solving the dependence on predetermined tasks in existing technologies. This allows the system to adaptively discover and respond to unexpected new tasks in the environment, greatly enhancing its adaptability to complex and dynamic environments. Subsequently, it receives and publishes intent reasoning conclusions fed back by the intent reasoning agent based on the aforementioned data, and task orchestration results fed back by the task orchestration agent based on the basic decision-making data and intent reasoning conclusions. Through this approach, an asynchronous, decoupled, data-driven information decision-making method based on a blackboard sharing network is constructed. Each agent, as an independent decision-making unit, independently performs information reading and writing, data perception, reasoning, and planning around the blackboard sharing network, realizing adaptive collaboration among multiple agents. Ultimately, without human intervention, it efficiently and flexibly responds to complex and dynamic task requirements, significantly improving the flexibility and execution efficiency of complex information decision-making task orchestration.

[0192] It should be noted that the information decision-making device provided in this application embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0193] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute the information decision-making methods provided in the above embodiments.

[0194] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0195] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the information decision-making methods provided in the above embodiments.

[0196] In another aspect, embodiments of this application also provide a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the information decision-making methods provided in the above embodiments.

[0197] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0198] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0199] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An information-based decision-making method, characterized in that, Applications in blackboard sharing networks include: Receive basic decision-making data sent by the perceptual agent, and publish the basic decision-making data; Receive the intent reasoning conclusion fed back by the intent reasoning agent based on the decision-making basis data, and publish the intent reasoning conclusion; Receive the task orchestration result fed back by the task orchestration agent based on the decision-making data and the intent reasoning conclusion; When a conflict is detected between the intent reasoning conclusion and the task orchestration result, a conflict event is issued; The system receives the arbitration result corresponding to the conflict event sent by the decision-making arbitration agent. The arbitration result includes the winning agent and the locking time. The arbitration result is determined as follows: Based on the conflict event, the system obtains the target conflicting agents and their current weights, the rationality score of the conflict content, and the time difference between the occurrence time of the conflict event and the current time; Based on the current weights, the rationality score, and the time difference, the system calculates the arbitration priority score of each target conflicting agent; The system selects the agent with the highest arbitration priority score as the winning agent, and the locking time is calculated based on the arbitration priority score of the winning agent. Based on the arbitration result, a lock announcement is issued. The lock announcement is used to indicate that the content of the winning agent is valid during the lock period and to instruct the unsuccessful agents to enter a backoff waiting state.

2. The information decision-making method according to claim 1, characterized in that, After issuing the lock announcement based on the arbitration result, the information decision-making method further includes: Obtain the content feedback score of the target conflict agent and obtain the average feedback score; The receiving feedback learning agent sends updated weights based on the content feedback score, the average feedback score, and the current weights; The updated weights are determined as follows: Calculate the difference between the content feedback score and the average feedback score; Using a reinforcement learning algorithm, the current weights of the winning agent are updated based on the difference.

3. The information decision-making method according to claim 1, characterized in that, The arbitration result also includes the backoff time of the unsuccessful agent, which is calculated based on the lock time, dynamic delay time and random perturbation value. The dynamic delay time is determined based on the number of historical failures of the unsuccessful agent. After issuing a lock announcement based on the arbitration result, wherein the lock announcement is used to indicate that the content of the winning agent is valid during the lock period and to instruct the unsuccessful agents to enter a backoff waiting state, the information decision-making method further includes: Receive the resubmission request triggered by the unsuccessful agent after the backoff time has elapsed; If no conflict is detected between the resubmitted requests, the corresponding operation is performed based on the resubmitted requests.

4. The information decision-making method according to any one of claims 1 to 3, characterized in that, Before receiving the task orchestration result from the task orchestration agent based on the decision-making foundation data and the intent reasoning conclusion, the information decision-making method further includes: Receive available device information sent by the device control agent, and publish the available device information; The task orchestration result received by the receiving task orchestration agent based on the decision-making data and the intent reasoning conclusion includes: Receive the task orchestration result fed back by the task orchestration agent based on the decision-making data, the intent reasoning conclusion, and the available device information; The task orchestration result is obtained by the task orchestration agent in the following manner: Based on the intent reasoning conclusion, the task to be executed is determined, and the task to be executed is decomposed to obtain the sub-tasks to be executed; Based on the decision-making data and the available device information, the execution priority score of the task to be executed is calculated; Based on the execution priority score, the available device information, and the sub-tasks to be executed, the tasks to be executed are sorted and resources are allocated to obtain the task orchestration result.

5. The information decision-making method according to claim 4, characterized in that, The information decision-making method also includes: Receive the task execution result and effect verification request fed back by the device control agent based on the task orchestration result, and publish the task execution result and the effect verification request; Receive effect verification data sent by the perceptual agent based on the effect verification request, and publish the effect verification data; The receiving feedback learning agent receives model parameters based on the effect verification data, the decision-making basis data, and the task execution results. These model parameters are used to calculate the arbitration priority score.

6. An information decision-making device, characterized in that, include: The basic data receiving module is used to receive and publish the basic decision data sent by the perceptual intelligent agent. The reasoning conclusion receiving module is used to receive the intention reasoning conclusion fed back by the intention reasoning agent based on the decision-making basis data, and to publish the intention reasoning conclusion; The orchestration result receiving module is used to receive the task orchestration result fed back by the task orchestration agent based on the decision-making basis data and the intent reasoning conclusion; The information decision-making device further includes a module for issuing a conflict event when a conflict is detected between the intention reasoning conclusion and the task arrangement result; The system receives the arbitration result corresponding to the conflict event sent by the decision arbitration agent. The arbitration result includes the winning agent and the locking time. The arbitration result is determined by obtaining the target conflict agent and its current weight, the rationality score of the conflict content, and the time difference between the occurrence time of the conflict event and the current time based on the conflict event. Based on the current weight, the rationality score, and the time difference, calculate the arbitration priority score for each of the target conflict agents; The agent with the highest arbitration priority score is selected as the winning agent, and the locking time is calculated based on the arbitration priority score of the winning agent. Based on the arbitration result, a lock announcement is issued. The lock announcement is used to indicate that the content of the winning agent is valid during the lock period and to instruct the unsuccessful agents to enter a backoff waiting state.

7. An information decision-making system, characterized in that, This includes a blackboard-sharing network, a perceptual agent, an intent-reasoning agent, a task orchestration agent, and a decision-arbitration agent; among which: The blackboard sharing network is used to execute the information decision-making method according to any one of claims 1 to 5; The sensing agent is used to collect basic decision-making data and send the basic decision-making data to the blackboard sharing network; The intent reasoning agent is used to obtain the decision-making basis data from the blackboard sharing network, determine the intent reasoning conclusion based on the decision-making basis data, and feed back the intent reasoning conclusion to the blackboard sharing network. The task orchestration agent is used to obtain the decision-making basis data and the intent reasoning conclusion from the blackboard sharing network, determine the task orchestration result based on the decision-making basis data and the intent reasoning conclusion, and feed back the task orchestration result to the blackboard sharing network. The blackboard sharing network is used to publish a conflict event when a conflict is detected between the intention reasoning conclusion and the task orchestration result; The decision-making arbitration agent is used to determine the arbitration result corresponding to the conflict event and send the arbitration result to the blackboard sharing network. The arbitration result includes the winning agent and the locking time. The arbitration result is determined as follows: based on the conflict event, the target conflicting agents and their current weights, the rationality score of the conflict content, and the time difference between the occurrence time of the conflict event and the current time are obtained; based on the current weights, the rationality score, and the time difference, the arbitration priority score of each target conflicting agent is calculated; the agent with the highest arbitration priority score is selected as the winning agent, and the locking time is calculated based on the arbitration priority score of the winning agent. The blackboard sharing network is also used to issue a lock announcement based on the arbitration result. The lock announcement is used to indicate that the content of the winning agent is valid during the lock time and to instruct the unsuccessful agents to enter a backoff waiting state.

8. The information decision-making system according to claim 7, characterized in that, The information decision-making system also includes an equipment control agent and a feedback learning agent; The device control agent is used to obtain the task orchestration results from the blackboard sharing network, execute the task, and send the task execution results and effect verification requests to the blackboard sharing network. The sensing agent is also used to collect effect verification data and send the effect verification data to the blackboard sharing network when it receives the effect verification request from the blackboard sharing network. The feedback learning agent is used to obtain the decision-making basis data, the task execution results, and the effect verification data from the blackboard sharing network, and to optimize the model parameters using reinforcement learning algorithms.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the information decision-making method according to any one of claims 1 to 5.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the information decision-making method according to any one of claims 1 to 5.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the information decision-making method according to any one of claims 1 to 5.

Images