Multi-agent collaborative decision support and optimization system

Abstract

This invention provides a multi-agent collaborative decision support and optimization system, belonging to the field of artificial intelligence technology. The system comprises a multi-agent cluster unit, a collaborative decision-making central unit, a distributed global optimization unit, and a closed-loop feedback iteration unit. These four units form a full-link closed-loop architecture through bidirectional communication links. This invention solves the technical problems of existing multi-agent systems, such as high single-point failure risk in centralized architectures, poor collaborative consistency in distributed architectures, fragmented decision-making and optimization processes, and insufficient robustness in dynamic scenarios. It achieves adaptive collaboration among heterogeneous agents, deep coupling of decision-making and optimization, and full-process closed-loop iterative optimization, significantly improving the decision-making accuracy, response speed, and operational stability of multi-agent systems in dynamic and uncertain environments. It can be widely applied in various scenarios such as smart industry, smart grids, vehicle-road cooperation, and unmanned swarm operations.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically to a multi-agent collaborative decision support and optimization system. Background Technology

[0002] Complex operational scenarios are increasingly trending towards clustered and distributed operation modes, and the application scope of multi-agent collaborative operations continues to expand. Existing similar operational architectures are mainly divided into two categories: centralized control mode and decentralized autonomous operation mode.

[0003] The centralized management and control model relies on a single central node to complete all task allocation decisions and data processing. The overall architecture is simple, but all running instructions and computing pressure are concentrated on a single node. The node's operating load is high for a long time, which can easily lead to running out or node failure. If a single node malfunctions, it will directly affect the normal operation of the entire equipment or cluster. The environmental adaptability is weak, and it is difficult to adapt to the needs of large-scale intelligent agent cluster expansion.

[0004] The decentralized autonomous operation mode delegates operational authority to individual agents, allowing each agent to independently complete local judgments and actions, resulting in faster response times and more flexible cluster expansion. However, this mode lacks a unified global coordination mechanism. Each agent operates only around its own local needs, leading to mismatched operational logic between agents, potential action conflicts and task overlaps, and difficulty in controlling the overall operational trajectory.

[0005] Meanwhile, in existing conventional systems, the decision-making and optimization / adjustment processes are independent and separate. Once a decision is made, it cannot be adaptively optimized and corrected based on actual on-site conditions, and the optimization / adjustment process cannot have a reverse effect on the initial task planning. The entire operational process lacks a continuous feedback and adjustment mechanism, making it impossible to make real-time adjustments based on changes in the on-site environment and fluctuations in execution status. Faced with dynamically changing and complex working conditions, the decision-making adaptability is insufficient, the overall operational coordination is low, and when heterogeneous intelligent agents are networked together, reasonable permission allocation and function matching cannot be achieved, further limiting the stable operation capability of the cluster.

[0006] To address this, a multi-agent collaborative decision support and optimization system is proposed. Summary of the Invention

[0007] The purpose of this invention is to overcome the above-mentioned defects in the prior art and provide a multi-agent collaborative decision support and optimization system. This system solves the technical problems of high single-point failure risk in centralized architecture, poor collaborative consistency in distributed architecture, fragmented decision-making and optimization processes, insufficient robustness in dynamic scenarios, and weak adaptability of heterogeneous agents in the prior art, thereby achieving efficient collaboration, accurate decision-making, and continuous optimization of the multi-agent system.

[0008] The specific technical solution is as follows: A multi-agent collaborative decision support and optimization system includes: a multi-agent cluster unit, a collaborative decision hub unit, a distributed global optimization unit, and a closed-loop feedback iteration unit; the collaborative decision hub unit is bidirectionally connected to the multi-agent cluster unit, the distributed global optimization unit, and the closed-loop feedback iteration unit, respectively, and the multi-agent cluster unit is bidirectionally connected to the distributed global optimization unit.

[0009] As a preferred embodiment of the present invention, the multi-agent cluster unit includes at least two heterogeneous agent sub-units, each of which has a built-in local decision-making module, a state perception module, and a data interaction module; the output end of the state perception module is connected to the input end of the local decision-making module, the input and output ends of the local decision-making module are bidirectionally connected to the data interaction module, and the data interaction module is bidirectionally connected to the collaborative decision-making central unit and the distributed global optimization unit, respectively.

[0010] In a preferred embodiment of the present invention, the collaborative decision-making central unit incorporates a global task decomposition module, an agent permission control module, a collaborative consistency verification module, and a central communication module. The output of the global task decomposition module is connected to the input of the agent permission control module, the output of the agent permission control module is connected to the input of the collaborative consistency verification module, the input and output of the collaborative consistency verification module are bidirectionally connected to the central communication module, and the central communication module is bidirectionally connected to the multi-agent cluster unit, the distributed global optimization unit, and the closed-loop feedback iteration unit, respectively.

[0011] As a preferred embodiment of the present invention, the distributed global optimization unit incorporates a parallel optimization solution module, a local-global target adaptation module, an optimization convergence control module, and an optimization communication module. The input and output terminals of the local-global target adaptation module are bidirectionally connected to the parallel optimization solution module and the optimization convergence control module, respectively. The input and output terminals of the parallel optimization solution module are bidirectionally connected to the optimization communication module. The optimization communication module is bidirectionally connected to the multi-agent cluster unit, the collaborative decision-making central unit, and the closed-loop feedback iteration unit, respectively.

[0012] In a preferred embodiment of the present invention, the closed-loop feedback iteration unit integrates an execution state evaluation module, a decision deviation identification module, an iterative strategy update module, and a feedback communication module. The output of the execution state evaluation module is connected to the input of the decision deviation identification module, the output of the decision deviation identification module is connected to the input of the iterative strategy update module, the input and output of the iterative strategy update module are bidirectionally connected to the feedback communication module, and the feedback communication module is bidirectionally connected to the multi-agent cluster unit, the collaborative decision-making central unit, and the distributed global optimization unit, respectively.

[0013] As a preferred embodiment of the present invention, the heterogeneous intelligent agent subunit includes a perception-type intelligent agent subunit, a decision-type intelligent agent subunit, an execution-type intelligent agent subunit, and a computing power-type intelligent agent subunit; the perception-type intelligent agent subunit is used to collect full-dimensional state data of the environment and the target, the decision-type intelligent agent subunit is used to generate specific decision-making schemes for the local scenario, the execution-type intelligent agent subunit is used to execute decision instructions and provide feedback on the execution status, and the computing power-type intelligent agent subunit is used to provide parallel computing power support for distributed optimization.

[0014] In a preferred embodiment of the present invention, the agent permission control module incorporates a confidence assessment submodule and a dynamic permission adjustment submodule. The input of the confidence assessment submodule receives agent state data, historical execution data, and environmental perception data uploaded by the multi-agent cluster unit, and its output is connected to the input of the dynamic permission adjustment submodule. The output of the dynamic permission adjustment submodule is connected to the local decision-making module of each heterogeneous agent subunit in the multi-agent cluster unit through a central communication module. The confidence assessment submodule is used to calculate the decision confidence of each heterogeneous agent subunit, and the dynamic permission adjustment submodule is used to dynamically adjust the local decision-making permission range and decision priority of the corresponding heterogeneous agent subunit based on the decision confidence.

[0015] In a preferred embodiment of the present invention, the parallel optimization solution module incorporates a multi-objective optimization solution submodule and a constraint dynamic adaptation submodule. The input of the multi-objective optimization solution submodule receives the global optimization objective issued by the collaborative decision-making central unit and the local optimization constraints uploaded by the multi-agent cluster unit. Its output is connected to the input of the constraint dynamic adaptation submodule, which in turn connects to the collaborative consistency verification module of the collaborative decision-making central unit and the local decision-making module of the multi-agent cluster unit via an optimization communication module. The multi-objective optimization solution submodule is used to simultaneously solve the global optimization objective and the local optimization objectives of each heterogeneous agent subunit. The constraint dynamic adaptation submodule is used to adjust the constraint boundaries of the optimization solution based on dynamic environmental changes and agent state updates.

[0016] In a preferred embodiment of the present invention, the output of the iterative strategy update module is connected to the global task decomposition module of the collaborative decision-making central unit, the optimization convergence control module of the distributed global optimization unit, and the local decision-making module of the multi-agent cluster unit, respectively. The iterative strategy update module generates a global task decomposition correction strategy, an optimization solution parameter update strategy, and a local decision boundary adjustment strategy based on the deviation data output by the decision deviation identification module, and distributes the corresponding strategies to the corresponding modules to realize a closed-loop iteration of decision generation, optimization solution, and execution feedback.

[0017] As a preferred embodiment of the present invention, the collaborative decision-making central unit, the multi-agent cluster unit, the distributed global optimization unit, and the closed-loop feedback iteration unit all adopt a low-latency, high-reliability bidirectional communication link, and the communication link has a built-in data encryption and time synchronization sub-module; the data encryption sub-module is used to perform end-to-end encryption on the interactive decision data, optimization data, and state data, and the time synchronization sub-module is used to achieve global time synchronization of all units and modules, ensuring the temporal consistency of collaborative decision-making and optimization iteration.

[0018] The present invention has the following beneficial effects: 1. By combining multiple units in a multi-unit deployment and employing a two-way interactive mode, the system integrates the advantages of centralized control and distributed autonomous operation. It no longer relies on a single central node to handle all computation and decision-making, thus distributing the overall system load, reducing the impact of local unit anomalies on the entire system, and improving the stability of the entire equipment cluster. Each unit has a clear division of labor and operates in close coordination with each other, preventing process disruptions caused by one-way data transmission and giving the overall architecture stronger fault tolerance.

[0019] 2. By establishing a hierarchical task decomposition and unified consistency verification mechanism, a large number of heterogeneous intelligent agents can be uniformly scheduled and managed. This mechanism can reasonably distinguish the functional positioning and operating permissions of different devices, orderly standardize various operational actions within the cluster, and reduce action conflicts and task duplication between intelligent agents. The dynamic permission adjustment mode can match the real-time operating status changes of devices, reasonably control the autonomous operating scope of different intelligent agents, and make the cluster collaborative operation more orderly and coordinated.

[0020] 3. It achieves a match between local operational needs and global operational goals, breaking down the disconnect between decision-making and optimization. Optimization results can correct decision content, and decision content can provide a complete basis for optimization calculations. Relying on a distributed parallel computing model, it fully utilizes idle computing resources within the cluster, adapts to complex multi-objective adjustment needs, dynamically adjusts operational constraints, fits changing field environments, and improves the adaptability and implementation effectiveness of various operational solutions.

[0021] 4. A complete closed-loop feedback and adjustment mechanism is added to track the entire operation process, enabling timely detection of various deviations during actual operation. Adjustments and corrections are simultaneously completed from multiple dimensions, including task planning, optimization, computation, and local decision-making. Relying on a cyclical iterative adjustment mode, the system can autonomously adapt to various dynamic scenarios such as environmental fluctuations and equipment status changes, continuously optimizing its operational logic and maintaining excellent long-term adaptability.

[0022] 5. A stable and reliable two-way communication mode ensures smooth and continuous data transmission across units, reducing coordination errors caused by information delays. A unified time base setting guarantees coordinated and consistent action sequences across all modules, while encrypted transmission enhances the security of data interaction. The categorized deployment of heterogeneous devices, coupled with flexible access adaptation capabilities, can accommodate different work scales and diverse application scenarios, making subsequent device expansion and scenario migration more convenient. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the connection relationship of the multi-agent collaborative decision support and optimization system of the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0025] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual images. They should not be construed as limiting the scope of this application. To better illustrate the embodiments of the present invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0026] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "inner," and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present application. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0027] In the description of this invention, unless otherwise explicitly specified and limited, the term "connection" or similar designation indicating a connection between components should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral part; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can refer to the internal communication between two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0028] Example The multi-agent collaborative decision support and optimization system provided in this embodiment, such as Figure 1As shown, it includes: a multi-agent cluster unit, a collaborative decision-making central unit, a distributed global optimization unit, and a closed-loop feedback iteration unit; the collaborative decision-making central unit is bidirectionally connected to the multi-agent cluster unit, the distributed global optimization unit, and the closed-loop feedback iteration unit, respectively, and the multi-agent cluster unit is bidirectionally connected to the distributed global optimization unit.

[0029] This solution establishes a complete system architecture through bidirectional communication connections between four units, enabling two-way data exchange between them and avoiding disruptions caused by unidirectional data flow. Each unit within the architecture can undertake the operational functions of its corresponding stage, eliminating the need for a single unit to bear the entire operational load and reducing the impact of a single unit failure on the entire system. Furthermore, the coordinated operation between the units ensures a complete workflow, guaranteeing smooth connections between each operational stage.

[0030] Specifically, in this embodiment, the multi-agent cluster unit includes at least two heterogeneous agent sub-units. Each heterogeneous agent sub-unit has a built-in local decision-making module, a state perception module, and a data interaction module. The output of the state perception module is connected to the input of the local decision-making module. The input and output of the local decision-making module are bidirectionally connected to the data interaction module. The data interaction module is bidirectionally connected to the collaborative decision-making central unit and the distributed global optimization unit, respectively.

[0031] In this scheme, the multi-agent cluster unit is configured with multiple heterogeneous intelligent agent sub-units, which can adapt to different operational functional requirements, allowing different intelligent agents to undertake operational tasks matching their own capabilities. The state perception module within each intelligent agent sub-unit can collect relevant data on its own operation and the surrounding environment in real time, providing stable data support for local decision-making; the local decision-making module can complete autonomous judgment and solution generation within its corresponding range, without the need for all decision content to be aggregated and transmitted upwards, reducing the time consumption caused by data flow; the data interaction module can ensure smooth data transmission between intelligent agents and other units, allowing local state and decision content to be transmitted outwards in a timely manner, while receiving external instructions and data, ensuring the coordinated operation of intelligent agents and other parts of the system.

[0032] Specifically, in this embodiment, the collaborative decision-making central unit has a built-in global task decomposition module, an agent permission control module, a collaborative consistency verification module, and a central communication module. The output of the global task decomposition module is connected to the input of the agent permission control module, the output of the agent permission control module is connected to the input of the collaborative consistency verification module, the input and output of the collaborative consistency verification module are bidirectionally connected to the central communication module, and the central communication module is bidirectionally connected to the multi-agent cluster unit, the distributed global optimization unit, and the closed-loop feedback iteration unit, respectively.

[0033] In this solution, the modules within the collaborative decision-making central unit are sequentially connected, enabling comprehensive management of the entire global task process. The global task decomposition module breaks down the top-level overall goal into actionable sub-tasks, allowing the overall goal to be assigned to various execution units, ensuring its feasibility. The agent permission control module regulates the operational permissions of each agent, ensuring different agents operate within their designated scopes and preventing system chaos caused by unauthorized operations. The collaborative consistency verification module verifies the decisions generated by each agent, preventing conflicts between their actions and ensuring all actions align with the overall goal. The central communication module ensures bidirectional data transmission between the central unit and other units, enabling timely issuance of instructions and real-time reception of status and data uploaded by other units, achieving overall coordination of the entire system from the central unit.

[0034] Specifically, in this embodiment, the distributed global optimization unit has a built-in parallel optimization solution module, a local-global target adaptation module, an optimization convergence control module, and an optimization communication module. The input and output terminals of the local-global target adaptation module are bidirectionally connected to the parallel optimization solution module and the optimization convergence control module, respectively. The input and output terminals of the parallel optimization solution module are bidirectionally connected to the optimization communication module. The optimization communication module is bidirectionally connected to the multi-agent cluster unit, the collaborative decision-making central unit, and the closed-loop feedback iteration unit, respectively.

[0035] In this scheme, the modules within the distributed global optimization unit work in two directions, enabling controllable operation of the entire optimization computation process. The local-global target adaptation module aligns the local operational goals of each agent with the overall system goal, preventing local actions from deviating from the overall direction. The parallel optimization solution module can simultaneously perform calculations on multiple optimization goals, eliminating the need for sequential processing in a single thread, thus shortening the computation time and fully utilizing the system's computing resources. The optimization convergence control module monitors and manages the entire optimization computation process, controlling the termination point to prevent infinite loops and avoids the optimization computation from falling into unreasonable result ranges. The optimization communication module ensures data exchange between the optimization unit and other units, allowing the optimization computation to obtain the necessary basic data in a timely manner and promptly transmitting the optimized results to the corresponding units, ensuring the optimization results can be applied practically.

[0036] Specifically, in this embodiment, the closed-loop feedback iteration unit integrates an execution status evaluation module, a decision deviation identification module, an iteration strategy update module, and a feedback communication module. The output of the execution status evaluation module is connected to the input of the decision deviation identification module, the output of the decision deviation identification module is connected to the input of the iteration strategy update module, the input and output of the iteration strategy update module are bidirectionally connected to the feedback communication module, and the feedback communication module is bidirectionally connected to the multi-agent cluster unit, the collaborative decision-making central unit, and the distributed global optimization unit, respectively.

[0037] In this scheme, the modules within the closed-loop feedback iteration unit are interconnected according to the process, enabling full-process tracking and adjustment of the system's operational status. The execution status assessment module can statistically evaluate the actual operation of each link within the system, accurately grasping the true state of the system's operation; the decision deviation identification module can compare the actual operating status with the expected target, accurately locate the deviation between the two, and clarify the specific degree and occurrence of the deviation; the iterative strategy update module can generate corresponding adjustment plans based on the identified deviations, correcting unreasonable parts in the operation process; the feedback communication module can promptly distribute the generated adjustment plans to the corresponding units and modules, ensuring that the adjustment plans are implemented and forming a complete closed loop in the operation of the entire system, achieving continuous adjustment and improvement of the operation process.

[0038] Specifically, in this embodiment, the heterogeneous intelligent agent subunit includes a perception-type intelligent agent subunit, a decision-type intelligent agent subunit, an execution-type intelligent agent subunit, and a computing power-type intelligent agent subunit. The perception-type intelligent agent subunit is used to collect full-dimensional state data of the environment and the target. The decision-type intelligent agent subunit is used to generate specific decision-making schemes for the local scenario. The execution-type intelligent agent subunit is used to execute decision instructions and provide feedback on the execution status. The computing power-type intelligent agent subunit is used to provide parallel computing power support for distributed optimization.

[0039] This solution clearly categorizes heterogeneous intelligent agent subunits by function, allowing different types of agents to specialize in functions matching their own attributes. This fully leverages the hardware and functional advantages of different agents, eliminating the need for a single agent to handle multiple mismatched tasks and improving the operational efficiency of each functional component. Furthermore, the categorized agents can be flexibly added or removed based on the needs of specific scenarios, enhancing the system's adaptability to different scenarios and its future expansion potential.

[0040] Specifically, in this embodiment, the agent permission control module has a built-in confidence assessment submodule and a permission dynamic adjustment submodule. The input end of the confidence assessment submodule receives agent state data, historical execution data, and environmental perception data uploaded by the multi-agent cluster unit, and the output end is connected to the input end of the permission dynamic adjustment submodule. The output end of the permission dynamic adjustment submodule is connected to the local decision-making module of each heterogeneous agent subunit in the multi-agent cluster unit through the central communication module. The confidence assessment submodule is used to calculate the decision confidence of each heterogeneous agent subunit, and the permission dynamic adjustment submodule is used to dynamically adjust the local decision-making permission range and decision priority of the corresponding heterogeneous agent subunit according to the decision confidence.

[0041] This solution incorporates two sub-modules within the agent permission management module, enabling fine-grained and dynamic management of agent permissions. The confidence assessment sub-module quantifies the reliability of an agent's operation based on various operational data uploaded by the agent, providing a reference for permission allocation. The dynamic permission adjustment sub-module adjusts the operational permission range and task priority of the corresponding agent in real time based on the assessment results. This allows stable and reliable agents to assume more autonomous operational permissions, while restricting the operational permissions of agents with abnormal states, avoiding system operational risks caused by agent anomalies. Furthermore, the dynamic adjustment method adapts to the real-time changes in the agent's operational state, improving the stability and flexibility of the system.

[0042] Specifically, in this embodiment, the parallel optimization solution module incorporates a multi-objective optimization solution submodule and a constraint dynamic adaptation submodule. The input of the multi-objective optimization solution submodule receives the global optimization objective issued by the collaborative decision-making central unit and the local optimization constraints uploaded by the multi-agent cluster unit. The output of the multi-objective optimization solution submodule is connected to the input of the constraint dynamic adaptation submodule. The output of the constraint dynamic adaptation submodule is connected to the collaborative consistency verification module of the collaborative decision-making central unit and the local decision-making module of the multi-agent cluster unit through the optimization communication module. The multi-objective optimization solution submodule is used to simultaneously solve the global optimization objective and the local optimization objectives of each heterogeneous agent subunit. The constraint dynamic adaptation submodule is used to adjust the constraint boundaries of the optimization solution according to the dynamic changes in the environment and the updates of the agent state.

[0043] This scheme incorporates two sub-modules within the parallel optimization solution module, enhancing the adaptability and practicality of the optimization solution. The multi-objective optimization solution sub-module can simultaneously handle the overall optimization objective of the system and the local optimization objectives of each agent, balancing the optimization needs of both the whole and the localities and avoiding operational imbalances in other aspects caused by focusing on only a single objective. The constraint dynamic adaptation sub-module can adjust the constraint boundaries of the optimization calculation in real time based on environmental changes and agent state updates, allowing the optimization calculation to adapt to real-time changing operating scenarios. This ensures that the optimization results do not lose their application value due to scenario changes, improving the usability of the optimization results in actual operation.

[0044] Specifically, in this embodiment, the output of the iterative strategy update module is connected to the global task decomposition module of the collaborative decision-making hub unit, the optimization convergence control module of the distributed global optimization unit, and the local decision-making module of the multi-agent cluster unit, respectively. The iterative strategy update module generates a global task decomposition correction strategy, an optimization solution parameter update strategy, and a local decision boundary adjustment strategy based on the deviation data output by the decision deviation identification module, and distributes the corresponding strategies to the corresponding modules to realize the full-link closed-loop iteration of decision generation, optimization solution, and execution feedback.

[0045] In this solution, the iterative strategy update module can directly interface with corresponding modules in multiple stages within the system. Based on the identified deviations, it generates corresponding adjustment strategies for the three stages: task decomposition, optimization calculation, and local decision-making. These adjustment strategies are then directly distributed to the corresponding execution modules, ensuring that deviations at different stages are specifically corrected and preventing incomplete corrections. Furthermore, it enables synchronous adjustments across the entire chain, from task planning and optimization calculation to local execution, guaranteeing consistent and coordinated adjustments throughout the system and allowing the closed-loop iteration to cover the entire system's operational flow.

[0046] Specifically, in this embodiment, the collaborative decision-making central unit, the multi-agent cluster unit, the distributed global optimization unit, and the closed-loop feedback iteration unit all adopt a low-latency, high-reliability bidirectional communication link. The communication link has a built-in data encryption and time synchronization sub-module. The data encryption sub-module is used to perform end-to-end encryption on the interactive decision data, optimization data, and state data. The time synchronization sub-module is used to achieve global time synchronization of all units and modules, ensuring the timing consistency of collaborative decision-making and optimization iteration.

[0047] This solution employs low-latency, highly reliable bidirectional communication links between system units, ensuring smooth data transmission, reducing transmission time and packet loss, and preventing action delays or collaborative failures due to data transmission issues. A data encryption submodule within the communication link encrypts various types of data during transmission, preventing data leakage or tampering and enhancing system security. A time synchronization submodule ensures all units and modules maintain a unified time base, preventing timing discrepancies and ensuring consistent timing of collaborative actions between units, thus improving the accuracy of system collaborative operation.

[0048] Working principle: This system relies on a multi-unit combined and linked overall architecture to build a complete operating logic of hierarchical collaborative distributed optimization closed-loop adjustment. It relies on bidirectional data communication between units to open up a complete operating link of task decomposition, decision generation, optimization calculation, action execution, feedback and correction.

[0049] In the initial stage of system operation, the collaborative decision-making central unit takes over the overall operational goals of the upper layer, breaks down the large-scale overall goals into subdivisions that are adapted to the carrying capacity of different intelligent agents, and combines the real-time operating status and historical performance of each intelligent agent to divide reasonable operating permissions and execution priorities, unify and standardize the overall operating direction of the cluster, and avoid conflicts caused by disorderly operation of individuals.

[0050] In a multi-agent cluster unit, various heterogeneous agents continuously perceive and collect information about the on-site environment, their own operating status, and the task objectives. Relying on local modules, they complete basic autonomous judgments and preliminary decision generation. At the same time, they upload the collected data and local operating plans to other units in real time, and synchronously receive task instructions and optimization adjustment information issued by the central unit to complete the execution of basic tasks.

[0051] The distributed global optimization unit synchronously receives global task requirements and local operating conditions of each agent, aligns local operating needs with overall operating goals, utilizes the distributed computing power of the cluster to carry out parallel computing, and adjusts constraints in real time in conjunction with changes in the external environment and equipment status. It performs overall optimization and adjustment of various operating schemes, corrects scheme deviations caused by a single local perspective, and ensures that the optimization results conform to global operating requirements.

[0052] The closed-loop feedback iterative unit tracks and monitors the actual execution process and implementation effects of each agent throughout the entire process. It compares the actual operation with the preset operation goals, identifies various deviations that occur during operation, distinguishes the links and scope of impact of the deviations, and outputs targeted adjustment strategies. The adjustment content is applied to the task decomposition logic optimization operation parameters and the agent's local decision boundary. Through continuous information feedback and strategy updates, the entire system's operation logic is cyclically corrected.

[0053] Each unit maintains consistent timing and secure data transmission throughout the process, ensuring a reasonable sequence of command issuance, data interaction, and strategy adjustments. Each module connects and cooperates according to a fixed logical order, with each layer working in tandem. This allows for coordinated global planning, decentralized execution, optimization, adjustment, feedback, and correction, ensuring the long-term stable and collaborative operation of the cluster.

[0054] How to use: Once the entire system is deployed, the networking connection of each intelligent agent device and the establishment of communication links between units are completed, global time calibration and data transmission security configuration are completed, and the normal communication interaction of each module is confirmed, it can be put into normal use.

[0055] In actual use, staff input the overall operation task and basic operation constraints into the collaborative decision-making central unit. The central unit automatically completes task breakdown and agent allocation, completes permission division based on the real-time status of each agent, and issues detailed tasks and operation rules to the multi-agent cluster unit.

[0056] Various types of intelligent agents continuously collect environmental and state perception data, generate local operational decisions based on their own permission scope, and simultaneously upload all collected data and decision content. The collaborative decision-making central unit performs consistency checks on all local decisions, eliminates conflicting operational plans, and completes the unified review of previous decisions.

[0057] The distributed global optimization unit receives decision content environmental data and task constraints in real time, performs multi-dimensional parallel optimization processing, adapts synchronously to dynamic changes in on-site conditions, outputs optimized and adjusted operation plans, and sends them back to the central unit and each intelligent agent to replace or correct the original decision content.

[0058] The intelligent agent carries out actual operations according to the optimized plan. During the operation, the closed-loop feedback iteration unit continuously collects the execution data of the entire process and regularly completes the operation status evaluation. Once a deviation between the actual execution and the target requirements is detected, the corresponding adjustment strategy is automatically generated and pushed to the corresponding functional units.

[0059] After receiving the adjustment strategy, each unit autonomously corrects its task division logic, optimizes computational conditions, and local decision-making rules, entering a new round of decision-making and execution processes. In daily use, the number of heterogeneous intelligent agents can be increased or decreased according to the scale of the operation. The system can autonomously complete the adaptation and permission allocation for newly added devices. During long-term operation, no frequent manual intervention is required; only periodic monitoring of the overall operating status is needed.

[0060] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-agent collaborative decision support and optimization system, characterized in that, include: Multi-agent cluster unit, collaborative decision-making central unit, distributed global optimization unit, and closed-loop feedback iteration unit; The collaborative decision-making central unit is bidirectionally connected to the multi-agent cluster unit, the distributed global optimization unit, and the closed-loop feedback iteration unit, respectively. The multi-agent cluster unit is bidirectionally connected to the distributed global optimization unit.

2. The multi-intelligence agent collaborative decision support and optimization system of claim 1, wherein, The multi-agent cluster unit includes at least two heterogeneous agent sub-units. Each heterogeneous agent sub-unit has a built-in local decision-making module, a state perception module, and a data interaction module. The output of the state perception module is connected to the input of the local decision-making module. The input and output of the local decision-making module are bidirectionally connected to the data interaction module. The data interaction module is bidirectionally connected to the collaborative decision-making central unit and the distributed global optimization unit, respectively.

3. The multi-intelligence agent collaborative decision support and optimization system of claim 1, wherein, The collaborative decision-making central unit has a built-in global task decomposition module, an agent permission control module, a collaborative consistency verification module, and a central communication module. The output of the global task decomposition module is connected to the input of the agent permission control module, the output of the agent permission control module is connected to the input of the collaborative consistency verification module, the input and output of the collaborative consistency verification module are bidirectionally connected to the central communication module, and the central communication module is bidirectionally connected to the multi-agent cluster unit, the distributed global optimization unit, and the closed-loop feedback iteration unit, respectively.

4. The multi-agent collaborative decision support and optimization system according to claim 1, characterized in that, The distributed global optimization unit incorporates a parallel optimization solution module, a local-global objective adaptation module, an optimization convergence control module, and an optimization communication module. The input and output terminals of the local-global objective adaptation module are bidirectionally connected to the parallel optimization solution module and the optimization convergence control module, respectively. The input and output terminals of the parallel optimization solution module are bidirectionally connected to the optimization communication module. The optimization communication module is bidirectionally connected to the multi-agent cluster unit, the collaborative decision-making central unit, and the closed-loop feedback iteration unit, respectively.

5. The multi-agent collaborative decision support and optimization system according to claim 1, characterized in that, The closed-loop feedback iteration unit integrates an execution status evaluation module, a decision deviation identification module, an iterative strategy update module, and a feedback communication module. The output of the execution status evaluation module is connected to the input of the decision deviation identification module, the output of the decision deviation identification module is connected to the input of the iterative strategy update module, the input and output of the iterative strategy update module are bidirectionally connected to the feedback communication module, and the feedback communication module is bidirectionally connected to the multi-agent cluster unit, the collaborative decision-making central unit, and the distributed global optimization unit, respectively.

6. The multi-agent collaborative decision support and optimization system according to claim 2, characterized in that, The heterogeneous intelligent agent subunits include perception-type intelligent agent subunits, decision-type intelligent agent subunits, execution-type intelligent agent subunits, and computing power-type intelligent agent subunits. The perception-type intelligent agent subunits are used to collect full-dimensional state data of the environment and the target. The decision-type intelligent agent subunits are used to generate specific decision-making schemes for local scenarios. The execution-type intelligent agent subunits are used to execute decision instructions and provide feedback on the execution status. The computing power-type intelligent agent subunits are used to provide parallel computing power support for distributed optimization.

7. The multi-agent collaborative decision support and optimization system according to claim 3, characterized in that, The intelligent agent permission control module has a built-in confidence evaluation submodule and a permission dynamic adjustment submodule. The input end of the confidence evaluation submodule receives intelligent agent status data, historical execution data and environmental perception data uploaded by the multi-agent cluster unit, and the output end is connected to the input end of the permission dynamic adjustment submodule. The output end of the permission dynamic adjustment submodule is connected to the local decision module of each heterogeneous intelligent agent subunit in the multi-agent cluster unit through the central communication module. The confidence assessment submodule is used to calculate the decision confidence of each heterogeneous intelligent agent subunit, and the dynamic permission adjustment submodule is used to dynamically adjust the local decision permission range and decision priority of the corresponding heterogeneous intelligent agent subunit according to the decision confidence.

8. The multi-agent collaborative decision support and optimization system according to claim 4, characterized in that, The parallel optimization solution module incorporates a multi-objective optimization solution submodule and a constraint dynamic adaptation submodule. The input of the multi-objective optimization solution submodule receives the global optimization objective from the collaborative decision-making central unit and the local optimization constraints uploaded by the multi-agent cluster unit. Its output is connected to the input of the constraint dynamic adaptation submodule, which in turn connects to the collaborative consistency verification module of the collaborative decision-making central unit and the local decision-making module of the multi-agent cluster unit via an optimization communication module. The multi-objective optimization solution submodule is used to simultaneously solve the global optimization objective and the local optimization objectives of each heterogeneous agent subunit. The constraint dynamic adaptation submodule is used to adjust the constraint boundaries of the optimization solution based on dynamic environmental changes and agent state updates.

9. The multi-agent cooperative decision support and optimization system according to claim 5, characterized in that, The output of the iterative strategy update module is connected to the global task decomposition module of the collaborative decision-making central unit, the optimization convergence control module of the distributed global optimization unit, and the local decision-making module of the multi-agent cluster unit. Based on the deviation data output by the decision deviation identification module, the iterative strategy update module generates a global task decomposition correction strategy, an optimization solution parameter update strategy, and a local decision boundary adjustment strategy, and distributes the corresponding strategies to the corresponding modules to realize a closed-loop iteration of decision generation, optimization solution, and execution feedback.

10. The multi-agent collaborative decision support and optimization system according to any one of claims 1-9, characterized in that, The collaborative decision-making central unit, multi-agent cluster unit, distributed global optimization unit, and closed-loop feedback iteration unit all employ low-latency, highly reliable bidirectional communication links. The communication links have built-in data encryption and time synchronization sub-modules. The data encryption sub-module is used to perform end-to-end encryption on the interactive decision data, optimization data, and state data. The time synchronization sub-module is used to achieve global time synchronization of all units and modules, ensuring the temporal consistency of collaborative decision-making and optimization iteration.

Images