A method and apparatus for heterogeneous management of intelligent agents based on a unified protocol

Abstract

This application discloses a method and apparatus for heterogeneous agent management based on a unified protocol, relating to the field of agent technology. The method includes: defining a unified protocol framework for agent communication in an enterprise's management server; registering several heterogeneous agents within the enterprise to obtain a heterogeneous agent resource library; and constructing an agent heterogeneous management model. Based on the unified protocol framework and the heterogeneous agent resource library, and according to user task information, using the agent heterogeneous management model, a scheduling scheme for the heterogeneous agents is generated and executed. Empirical data after the scheduling scheme execution is collected, and based on this data, the agent heterogeneous management model is self-evolved to obtain an updated model, which then waits to receive the next task information. This method solves the problems of communication and integration difficulties, poor agent scheduling performance, and low intelligence level in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of intelligent agent technology, and in particular to a method and apparatus for heterogeneous management of intelligent agents based on a unified protocol. Background Technology

[0002] With the development of artificial intelligence technology, heterogeneous intelligent agent architectures, composed of intelligent agents with different algorithms, architectures, and functions, have been widely used in fields such as intelligent manufacturing, smart logistics, and smart cities. Enterprise management servers often contain multiple heterogeneous intelligent agents to perform different tasks across different production lines and departments. However, effectively managing and coordinating these heterogeneous intelligent agents to form an efficient and unified whole remains a major technical challenge.

[0003] The existing technology has the following main drawbacks:

[0004] 1) Communication and Integration Difficulties: Traditional methods often employ point-to-point adapters or simple message brokers for integration. Point-to-point methods have poor scalability; each new agent type requires the development of a new adapter, resulting in high maintenance costs. Simple message brokers only solve the "transmission" problem, failing to address semantic uniformity. The diverse message formats and data meanings sent by different agents make it difficult for the management server to understand the true intent behind the messages, hindering deep, intentionally clear collaboration and creating "protocol silos" and a "semantic gap."

[0005] 2) Poor agent scheduling performance: Existing agent scheduling algorithms are mostly based on static rules or simple heuristics (such as round-robin or least-connection algorithms). These methods can work in stable environments and predictable tasks, but in dynamic, complex, and ever-changing real-world application environments (such as dynamic changes in task priorities, sudden agent failures, and network latency fluctuations), their response is slow, and the scheduling results often deviate from the global optimum. The lack of predictive ability for the future state of the system and the ability to globally weigh complex constraints leads to low resource utilization, prolonged task completion time, and poor system robustness.

[0006] 3) Low level of intelligence: Existing technologies generally lack the ability to learn and evolve. Behavioral logic is essentially fixed at deployment time, unable to learn from historical execution data, let alone self-adjust and optimize according to environmental changes. When the system scales up or business processes change, manual reconfiguration and adjustment are often required, resulting in low intelligence and difficulty in adapting to future development needs. Summary of the Invention

[0007] This invention provides a method and apparatus for heterogeneous management of intelligent agents based on a unified protocol, which solves the problems of communication and integration difficulties, poor intelligent agent scheduling effect and low level of intelligence in the prior art.

[0008] In a first aspect, embodiments of the present invention provide a method for heterogeneous management of intelligent agents based on a unified protocol, the method comprising:

[0009] In the enterprise's management server, a unified protocol framework for agent communication is defined, several heterogeneous agents of the enterprise are registered, a heterogeneous agent resource library is obtained, and an agent heterogeneous management model is constructed.

[0010] Based on a unified protocol framework and a heterogeneous agent resource library, and according to the user's task information, a heterogeneous agent management model is used to generate a scheduling scheme for heterogeneous agents and execute the scheduling scheme.

[0011] Collect experience data after the execution of the scheduling scheme, and perform self-evolution training on the agent heterogeneous management model based on the experience data to obtain an updated agent heterogeneous management model, and wait to receive the next task information.

[0012] The technical solution provided in this application has at least the following beneficial effects:

[0013] Breaking down protocol silos and achieving semantic-level unified collaboration, this project fundamentally solves the communication and integration challenges between heterogeneous intelligent agents by defining a unified protocol framework that includes structured message formats, communication interfaces, and interaction flows. It not only enables message transmission but, more importantly, unifies message semantics, allowing the management server to accurately understand the intentions and capabilities of each agent. This lays a solid foundation for deep collaboration and possesses extremely high scalability and maintainability. Furthermore, it achieves forward-looking, globally optimal intelligent scheduling by combining deep learning prediction, swarm intelligence optimization, and reinforcement learning. The agent resource state prediction model can predict future resource states, and the generated dynamic resource state prediction matrix serves as the core input, driving the scheduling scheme generation model to perform global optimization. The initial plan is then fine-tuned to adapt to real-time changes. This closed-loop mechanism of "prediction-optimization-re-optimization" makes scheduling decisions no longer a passive response based on current static information, but an active planning based on future dynamic trends. This significantly improves resource utilization, task completion efficiency, and system robustness. A complete self-evolving closed loop is constructed. By collecting experience data after scheduling execution and using the experience replay pool to continuously train the management model, it can learn from historical data and continuously optimize prediction accuracy, scheduling strategies, and anomaly detection capabilities. This gives the entire heterogeneous management architecture of the intelligent agent self-evolving characteristics, enabling it to adapt to environmental changes and business expansion, greatly reducing manual maintenance costs, and improving the level of intelligence and long-term value.

[0014] In one alternative implementation, a unified protocol framework for agent communication is defined in the enterprise's management server. An agent heterogeneous management model is constructed, and several heterogeneous agents within the enterprise are registered to obtain a heterogeneous agent resource library, including:

[0015] In the enterprise's management server, a structured message format, communication interface, interaction process, and protocol stack are defined to obtain a unified protocol framework for intelligent agent communication.

[0016] Based on a unified protocol framework, dynamic registration and capability modeling are performed on several heterogeneous intelligent agents in the management server to obtain a heterogeneous intelligent agent resource library, and a heartbeat mechanism is set for each intelligent agent.

[0017] Based on a unified protocol framework, an intelligent agent heterogeneous management model is constructed using artificial intelligence algorithms, and the intelligent agent heterogeneous management model is connected to a heterogeneous intelligent agent resource library.

[0018] In one alternative implementation, the message format includes a message header, a message body, and metadata;

[0019] The message header includes the message ID, source agent ID, target agent, timestamp, and message type;

[0020] The message body includes the actual data for the current message type;

[0021] Metadata includes the QoS requirements and security credentials for the current message body.

[0022] In one optional implementation, the message types include registration message types, heartbeat message types, task request message types, task instruction message types, status report message types, and business data message types.

[0023] In one optional implementation, based on a unified protocol framework, several heterogeneous agents in the management server are dynamically registered and their capabilities are modeled to obtain a heterogeneous agent resource library. A heartbeat mechanism is then set for each agent, including:

[0024] Based on a unified protocol framework, the system collects registration messages from several heterogeneous pre-set and / or new intelligent agents in the management server, and performs security verification on the identity credentials of the intelligent agents based on the registration messages.

[0025] If the security verification of the current agent passes, proceed to the next step; otherwise, delete the corresponding agent, issue an alarm signal, and terminate the registration of the corresponding agent.

[0026] Based on the registration message, the corresponding capability list is parsed, the capability list is converted into the corresponding intelligent agent capability model, and the intelligent agent capability model is stored in the heterogeneous intelligent agent resource library.

[0027] By traversing all agents, a heterogeneous agent resource library including the capability models of all agents is obtained, and a heartbeat mechanism is set for each agent, sending heartbeat messages to the management server based on a unified protocol framework.

[0028] In one alternative implementation, the agent heterogeneous management model includes an agent resource state prediction model based on deep learning algorithms, a scheduling scheme generation model based on swarm intelligence optimization algorithms, a scheduling scheme optimization model based on reinforcement learning algorithms, and an agent anomaly detection model based on deep learning algorithms.

[0029] In one optional implementation, the agent resource state prediction model is constructed based on the GNN-LSTM-MLP algorithm, and the agent resource state prediction model includes an agent communication relationship feature extraction module constructed based on the GNN algorithm, an agent state temporal feature extraction module constructed based on the LSTM algorithm, and an agent resource state prediction module constructed based on the MLP algorithm.

[0030] The scheduling scheme generation model is built based on the IBOA algorithm and includes an initialization module, an iterative optimization module, and an optimal decoding module.

[0031] The scheduling scheme optimization model is built based on the MPO-PPO algorithm, and includes a policy network optimization module based on the MPO algorithm and a scheduling scheme optimization module based on the PPO algorithm. The scheduling scheme optimization module is equipped with an experience replay pool.

[0032] The agent anomaly detection model is built based on the RF-Attention-SVM algorithm, and includes a key feature extraction module based on the RF algorithm, a weighted fusion module based on the Attention mechanism, and an agent anomaly detection module based on the SVM algorithm.

[0033] In one optional implementation, based on a unified protocol framework and a heterogeneous agent resource library, a heterogeneous agent scheduling scheme is generated using an agent heterogeneous management model according to the user's task information, and the scheduling scheme is executed, including:

[0034] It receives task information input by the user, converts the task information into a task request message based on a unified protocol framework, and transmits the task request message to the management server.

[0035] The system invokes a pre-defined domain knowledge graph to decompose the task request messages within the management server, thereby obtaining at least one task capability requirement.

[0036] Based on the task capability requirements, query the heterogeneous intelligent agent resource library, filter out all candidate intelligent agents that meet the requirements, and obtain the corresponding candidate intelligent agent pool.

[0037] Based on a unified protocol framework, the status report messages of each candidate agent in the candidate agent pool are collected, and several status report messages are input into the agent resource status prediction model in the agent heterogeneous management model.

[0038] Based on several status report messages, the agent resource status prediction model is used to generate a dynamic resource status prediction matrix of task capability requirements, and input into the scheduling scheme generation model in the agent heterogeneous management model.

[0039] Based on the dynamic resource state prediction matrix, the scheduling scheme generation model is used to generate an initial scheduling scheme for heterogeneous agents with task capability requirements, and input it into the scheduling scheme optimization model in the agent heterogeneous management model.

[0040] Based on the dynamic resource state prediction matrix and the initial scheduling scheme, the scheduling scheme optimization model is used to optimize the initial scheduling scheme and generate the final scheduling scheme of heterogeneous agents with task capability requirements.

[0041] Based on a unified protocol framework and according to the final scheduling scheme, task instruction messages for the target intelligent agents that require task capabilities are generated, and the task instruction messages are sent to the corresponding target intelligent agents.

[0042] Using the target intelligent agent, receive task instruction messages, and send business data messages to other target intelligent agents with the same task capability requirements based on a unified protocol framework;

[0043] Based on business data messages, all target agents with the same task capability requirements synchronously execute the task instruction messages of the final scheduling scheme, and during execution, each target agent sends the latest status report message to the management server.

[0044] Based on the latest status report, the agent anomaly detection module is used to generate the agent anomaly detection result for the target agent. If the agent anomaly detection result indicates that an anomaly exists, rescheduling is triggered, and the process returns to the agent resource status prediction step.

[0045] In one optional implementation, empirical data after the execution of the scheduling scheme is collected, and based on this empirical data, the agent heterogeneous management model is self-evolved and trained to obtain an updated agent heterogeneous management model, which then waits to receive the next task information, including:

[0046] Collect experience data after the execution of the scheduling scheme, store the experience data in the experience replay pool of the scheduling scheme optimization model of the agent heterogeneous management model, and periodically extract a number of experience data from the experience replay pool.

[0047] Based on some empirical data, the agent heterogeneous management model is trained in a cache area through self-evolution, and the model update amount of the updated agent heterogeneous management model is extracted.

[0048] If the model update amount exceeds the model update amount threshold, the agent heterogeneous management model in the memory area of ​​the management server will be adjusted, the corresponding updated agent heterogeneous management model will be stored, and the server will wait to receive the next task information.

[0049] Secondly, embodiments of the present invention provide an intelligent agent heterogeneous management device based on a unified protocol, used to implement an intelligent agent heterogeneous management method, the device comprising:

[0050] The initialization unit is used to define a unified protocol framework for agent communication in the enterprise's management server, register several heterogeneous agents of the enterprise, obtain a heterogeneous agent resource library, and build a heterogeneous agent management model.

[0051] The agent heterogeneous management unit is used to generate a scheduling scheme for heterogeneous agents based on a unified protocol framework and a heterogeneous agent resource library, according to the user's task information, using the agent heterogeneous management model, and then execute the scheduling scheme.

[0052] The self-evolutionary training unit is used to collect experience data after the execution of the scheduling scheme, and to perform self-evolutionary training on the agent heterogeneous management model based on the experience data to obtain an updated agent heterogeneous management model, and wait to receive the next task information.

[0053] A third aspect of this invention provides an electronic device, which includes:

[0054] At least one processor; and a memory communicatively connected to the at least one processor; wherein,

[0055] The memory stores instructions that can be executed by at least one processor, such that the at least one processor can perform the method proposed in the first aspect of the present invention.

[0056] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0057] Figure 1This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention;

[0058] Figure 2 This is a flowchart illustrating the steps of a heterogeneous management method for intelligent agents based on a unified protocol, as provided in an embodiment of the present invention.

[0059] Figure 3 This is a functional unit diagram of a heterogeneous management device for intelligent agents based on a unified protocol, provided in an embodiment of the present invention. Detailed Implementation

[0060] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0061] The present invention will be further described below with reference to the accompanying drawings.

[0062] Reference Figure 1 , Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0063] like Figure 1 As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0064] Those skilled in the art will understand that Figure 1The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0065] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating device, a data storage module, a network communication module, a user interface module, and electronic programs.

[0066] exist Figure 1 In the electronic device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the intelligent agent heterogeneous management device based on the unified protocol stored in the memory 1005 through the processor 1001 and executes the intelligent agent heterogeneous management method based on the unified protocol provided in the embodiment of the present invention.

[0067] Reference Figure 2 The present invention provides an intelligent content recommendation method based on an artificial intelligence (AI) model, the method comprising:

[0068] S201: In the enterprise's management server, define a unified protocol framework for agent communication, register several heterogeneous agents of the enterprise, obtain a heterogeneous agent resource library, and build an agent heterogeneous management model.

[0069] S202: Based on a unified protocol framework and a heterogeneous agent resource library, and according to the user's task information, a heterogeneous agent management model is used to generate a scheduling scheme for heterogeneous agents and execute the scheduling scheme.

[0070] S203: Collect experience data after the execution of the scheduling scheme, and perform self-evolution training on the agent heterogeneous management model based on the experience data to obtain an updated agent heterogeneous management model, and wait to receive the next task information.

[0071] The technical solution provided in this application has at least the following beneficial effects:

[0072] Breaking down protocol silos and achieving semantic-level unified collaboration, this project fundamentally solves the communication and integration challenges between heterogeneous intelligent agents by defining a unified protocol framework that includes structured message formats, communication interfaces, and interaction flows. It not only enables message transmission but, more importantly, unifies message semantics, allowing the management server to accurately understand the intentions and capabilities of each agent. This lays a solid foundation for deep collaboration and possesses extremely high scalability and maintainability. Furthermore, it achieves forward-looking, globally optimal intelligent scheduling by combining deep learning prediction, swarm intelligence optimization, and reinforcement learning. The agent resource state prediction model can predict future resource states, and the generated dynamic resource state prediction matrix serves as the core input, driving the scheduling scheme generation model to perform global optimization. The initial plan is then fine-tuned to adapt to real-time changes. This closed-loop mechanism of "prediction-optimization-re-optimization" makes scheduling decisions no longer a passive response based on current static information, but an active planning based on future dynamic trends. This significantly improves resource utilization, task completion efficiency, and system robustness. A complete self-evolving closed loop is constructed. By collecting experience data after scheduling execution and using the experience replay pool to continuously train the management model, it can learn from historical data and continuously optimize prediction accuracy, scheduling strategies, and anomaly detection capabilities. This gives the entire heterogeneous management architecture of the intelligent agent self-evolving characteristics, enabling it to adapt to environmental changes and business expansion, greatly reducing manual maintenance costs, and improving the level of intelligence and long-term value.

[0073] In one alternative implementation, a unified protocol framework for agent communication is defined in the enterprise's management server. An agent heterogeneous management model is constructed, and several heterogeneous agents within the enterprise are registered to obtain a heterogeneous agent resource library, including:

[0074] S2011: In the enterprise's management server, define structured message formats, communication interfaces, interaction processes, and protocol stacks to obtain a unified protocol framework for intelligent agent communication.

[0075] In this embodiment, the message format includes a message header, a message body, and metadata;

[0076] The message header includes the message ID (globally unique identifier), source agent ID, target agent, timestamp, and message type (which defines the purpose of the message).

[0077] Message types include registration message types, heartbeat message types, task request message types, task instruction message types, status report message types, and business data message types;

[0078] Registration message type: When a new intelligent agent (such as a robot, a data analysis module, or a sensor network) comes online, it needs to "report" to the central management server. This message is equivalent to the intelligent agent's "self-introduction" and "capability list", letting the manager know its existence, identity, and what it can do;

[0079] Heartbeat message type: After successful registration, the agent needs to periodically send heartbeat messages to the management server. This has two main functions: 1) to inform the manager that "I am still alive and in normal condition"; 2) to maintain an active session connection. If the manager does not receive a heartbeat from an agent within a preset time (such as 30 seconds), it will mark the agent as "offline" or "abnormal" and trigger the corresponding alarm or fault tolerance mechanism.

[0080] Task request message type: When a user needs to perform a service, they send a task request to the management server. After receiving the request, the management server will perform intelligent scheduling based on the request content, the current status and load of all agents, and select the most suitable agent to perform the task.

[0081] Task instruction message type: includes specific task parameters, execution requirements, etc., used to tell the agent which part of the scheduling scheme to execute and how to execute the scheduling scheme;

[0082] Status report message type; This is a message from the agent to the manager to report its "detailed health check report", which may include current CPU utilization, memory usage, length of pending task queue, network latency, etc. Unlike heartbeat, status reports are not sent periodically, but are actively reported when the status changes significantly, or sent when the management server actively queries, providing the most critical real-time data input for the management server's intelligent scheduling and load balancing.

[0083] Business data message type; used for transmitting actual business data between intelligent agents. This is the information carrier that carries the final value. For example, a visual sensor intelligent agent sends the image data it recognizes to a data analysis intelligent agent.

[0084] The message body includes the actual data for the current message type;

[0085] Metadata includes the current message body's Quality of Service (QoS) requirements (such as real-time performance and reliability levels) and security credentials (such as digital signatures and access tokens) to ensure the security and quality of communication.

[0086] The communication interface defines the standard communication interface (such as RESTful API, gRPC, MQTT Topic) between the agent and the management server, as well as between agents.

[0087] The interaction process definition includes the registration process, which requires the agent to first send a registration message, and the management server to verify and return a receiving or rejecting message.

[0088] A protocol stack is defined as a lightweight protocol stack software module developed or integrated on the management server side and each intelligent agent that needs to access the system, responsible for message encapsulation, parsing, sending and receiving.

[0089] S2012: Based on a unified protocol framework, dynamic registration and capability modeling are performed on several heterogeneous intelligent agents in the management server to obtain a heterogeneous intelligent agent resource library, and a heartbeat mechanism is set for each intelligent agent.

[0090] S2013: Based on the unified protocol framework, use artificial intelligence algorithms to construct a heterogeneous management model for intelligent agents, and connect the heterogeneous management model for intelligent agents to a heterogeneous intelligent agent resource library.

[0091] In one optional implementation, based on a unified protocol framework, several heterogeneous agents in the management server are dynamically registered and their capabilities are modeled to obtain a heterogeneous agent resource library. A heartbeat mechanism is then set for each agent, including:

[0092] S20121: Based on a unified protocol framework, collect registration messages from several heterogeneous preset and / or new intelligent agents in the management server, and perform security verification on the identity credentials of the intelligent agents according to the registration messages.

[0093] In this embodiment, after a new and / or pre-defined intelligent agent comes online, its built-in protocol stack module automatically sends a registration message conforming to a unified protocol to the management server. The message body contains a self-describing list of capabilities, for example:

[0094] For Automated Guided Vehicle (AGV) trolleys: {"type": "AGV","max_speed": "1.5m / s", "payload": "50kg", "navigation": "LiDAR", "battery_level": "95%"};

[0095] For visual inspection software: {"type": "Vision_Analyzer", "model": "YOLOv8", "input_format": "image / jpeg", "processing_fps": "30", "accuracy": "99.2%"};

[0096] S20122: If the security verification of the current agent passes, proceed to the next step; otherwise, delete the corresponding agent, issue an alarm signal, and end the registration of the corresponding agent.

[0097] S20123: Based on the registration message, parse the corresponding capability list, convert the capability list into the corresponding agent capability model, and store the agent's ID, IP address, agent capability model, current status (idle, busy, offline) and other information into the heterogeneous agent resource library;

[0098] In this embodiment, the agent capability model is a multi-dimensional vector, for example, AGV_Agent = [type_AGV,1.5, 50, 1, 0.95, ...], where each dimension corresponds to a quantified capability index;

[0099] S20124: Traverse all agents to obtain a heterogeneous agent resource library including the capability models of all agents, and set a heartbeat mechanism for each agent, and send heartbeat messages to the management server based on a unified protocol framework;

[0100] In this embodiment, after successful registration, the agent begins to send heartbeat messages to the management server at fixed intervals (e.g., 30 seconds). If the management server does not receive a heartbeat from an agent within three consecutive intervals, it marks the agent's status as "offline" and triggers updates to the heterogeneous agent resource library and potential alarms or task rescheduling.

[0101] In one alternative implementation, the agent heterogeneous management model includes an agent resource state prediction model based on deep learning algorithms, a scheduling scheme generation model based on swarm intelligence optimization algorithms, a scheduling scheme optimization model based on reinforcement learning algorithms, and an agent anomaly detection model based on deep learning algorithms.

[0102] In one alternative implementation, the agent resource state prediction model is constructed based on the Graph Neural Network (GNN)-Long Short-Term Memory (LSTM)-Multi-Layer Perceptron (MLP) algorithm, and the agent resource state prediction model includes an agent communication relationship feature extraction module constructed based on the GNN algorithm, an agent state temporal feature extraction module constructed based on the LSTM algorithm, and an agent resource state prediction module constructed based on the MLP algorithm.

[0103] In this embodiment, the agent communication relationship feature extraction module models the agents and their communication relationships as an agent communication graph, where agents are nodes and communication links are edges. The GNN is used to extract the topological relationships and dependency features between agents.

[0104] Agent state temporal feature extraction module: Receives historical state temporal data (such as CPU load in the past hour) of each agent (node) and extracts the temporal features of its state changes;

[0105] Agent resource state prediction module: It integrates the relational features output by GNN and the temporal features output by LSTM, and finally predicts the resource state (such as the predicted values ​​of CPU, memory and network bandwidth) of each agent within a future period (such as the next 30 minutes). It outputs a dynamic resource state prediction matrix. The rows of the matrix are agent IDs, the columns are the prediction time points and resource types, and the values ​​in the matrix are the predicted resource state values.

[0106] The scheduling scheme generation model is built on the Improved Bull Optimization Algorithm (IBOA) and includes an initialization module, an iterative optimization module, and an optimal decoding module.

[0107] In this embodiment, the initialization module encodes the scheduling scheme into individual vectors of the IBOA algorithm and generates a set of initial IBOA individuals based on task requirements and the candidate agent pool.

[0108] Iterative optimization module: This module needs to clearly define the objectives of generating the scheduling scheme. These objectives are usually multifaceted, for example:

[0109] Minimize total task completion time: Complete all tasks as early as possible, and set task delay costs;

[0110] Minimize load imbalance: Avoid some agents being overloaded while others are idle, and set resource overload costs;

[0111] Minimize communication overhead: Schedule tasks that require frequent communication to agents with low network latency and high bandwidth, and set communication costs;

[0112] In each iteration, a fitness function is constructed based on the target generated by the scheduling scheme. The cost prediction values ​​for all fitness functions are obtained from the dynamic resource state prediction matrix. Through continuous iteration and updating, the optimal solution is output.

[0113] Optimal solution decoding module: When the iteration reaches the termination condition (such as the maximum number of iterations or fitness convergence), the currently found optimal solution (i.e. the optimal scheduling scheme) is decoded into an executable instruction sequence to form the initial scheduling scheme;

[0114] The scheduling scheme optimization model is built based on the Meta-Policy Optimization (MPO) - Proximal Policy Optimization (PPO) algorithm. The scheduling scheme optimization model includes a policy network optimization module based on the MPO algorithm and a scheduling scheme optimization module based on the PPO algorithm. The scheduling scheme optimization module is equipped with an experience replay pool.

[0115] In this embodiment, the policy network optimization module is used to optimize the policy network of the scheduling scheme optimization module, so that it can achieve a better balance between exploring new schemes and utilizing known good schemes, which is especially suitable for continuous control spaces with complex constraints.

[0116] The scheduling scheme optimization module, as the main optimization engine, receives the initial scheduling scheme and dynamic resource state prediction matrix as state inputs, and takes fine-tuning of the scheduling scheme (such as adjusting the weight of task allocation and adjusting the execution order) as actions. Its goal is to learn a policy so that the fine-tuned scheme can obtain higher cumulative rewards (the reward function is similar to the fitness function, but focuses more on real-time feedback);

[0117] Experience replay pool: Stores a large amount of experience data (state, action, reward, next state) to break data correlation and improve training stability and efficiency;

[0118] The agent anomaly detection model is built based on the Random Forest (RF)-Attention-Support Vector Machine (SVM) algorithm. The agent anomaly detection model includes a key feature extraction module based on the RF algorithm, a weighted fusion module based on the Attention mechanism, and an agent anomaly detection module based on the SVM algorithm.

[0119] Key feature extraction module: used to filter out the key features most sensitive to anomalies from the high-dimensional state data reported by the agent, reducing data dimensionality and noise;

[0120] Weighted fusion module: performs weighted fusion on the selected key features and automatically learns the importance weight of different features for anomaly detection in different contexts;

[0121] The agent anomaly detection module acts as a classifier, receiving the weighted and fused feature vector and outputting a predicted label indicating whether the agent's state is "normal" or "abnormal".

[0122] In one optional implementation, based on a unified protocol framework and a heterogeneous agent resource library, a heterogeneous agent scheduling scheme is generated using an agent heterogeneous management model according to the user's task information, and the scheduling scheme is executed, including:

[0123] S2021: Receive task information input by the user, convert the task information into a task request message based on the unified protocol framework, and transmit the task request message to the management server;

[0124] In this embodiment, the management server receives a high-level task message via API or user interface, such as: "Move a batch of materials from warehouse A to workstation B on the production line".

[0125] S2022: Call the preset domain knowledge graph to decompose the task request message inside the management server and obtain at least one task capability requirement;

[0126] In this embodiment, the domain knowledge graph defines the task type, the required atomic capabilities, and the temporal and logical relationships between capabilities. For example, the "transportation task" is decomposed in the knowledge graph into: [perceive material location - plan path - grab material - transport - release material - report completion]. Each task capability requirement (such as grabbing material) is associated with the required capability model features (such as requiring an agent with type robotic_arm and payload > 20kg).

[0127] S2023: Based on the task capability requirements, query the heterogeneous intelligent agent resource library, filter out all candidate intelligent agents that meet the requirements, and obtain the corresponding candidate intelligent agent pool.

[0128] S2024: Based on a unified protocol framework, collect the status report messages of each candidate agent in the candidate agent pool, and input several status report messages into the agent resource status prediction model in the agent heterogeneous management model.

[0129] S2025: Based on several status report messages, use the agent resource status prediction model to generate a dynamic resource status prediction matrix of task capability requirements, and input it into the scheduling scheme generation model in the agent heterogeneous management model.

[0130] In this embodiment, the matrix quantifies the resource status and expected cost of assigning a subtask to a candidate agent at different points in the future;

[0131] S2026: Based on the dynamic resource state prediction matrix, use the scheduling scheme generation model to generate an initial scheduling scheme for heterogeneous agents with task capability requirements, and input it into the scheduling scheme optimization model in the agent heterogeneous management model.

[0132] S2027: Based on the dynamic resource state prediction matrix and the initial scheduling scheme, the scheduling scheme optimization model is used to optimize the initial scheduling scheme and generate the final scheduling scheme of heterogeneous agents with task capability requirements.

[0133] S2028: Based on the unified protocol framework and according to the final scheduling scheme, generate task instruction messages for the target intelligent agent that require task capabilities, and send the task instruction messages to the corresponding target intelligent agent.

[0134] S2029: Using the target intelligent agent, receive task instruction messages and, based on a unified protocol framework, send business data messages to other target intelligent agents with the same task capability requirements;

[0135] S20210: Based on the business data message, all target agents using the same task capability requirements synchronously execute the task instruction message of the final scheduling scheme, and during the execution, each target agent sends the latest status report message to the management server.

[0136] S20211: Based on the latest status report message, use the agent anomaly detection module to generate the agent anomaly detection result for the target agent. If the agent anomaly detection result indicates that an anomaly exists, trigger rescheduling and return to the agent resource status prediction step.

[0137] In one optional implementation, based on several status report messages, a dynamic resource status prediction matrix of task capability requirements is generated using an agent resource status prediction model, and then input into a scheduling scheme generation model within the agent heterogeneous management model, including:

[0138] S20251: The agent communication relationship feature extraction module uses the agent resource state prediction model to construct an agent communication graph, where agents are nodes, communication links are edges, and state report messages are used as node features.

[0139] S20252: Extract the agent communication relationship features from the agent communication graph;

[0140] S20253: Agent state temporal feature extraction module using agent resource state prediction model, extracts agent state temporal features of node features;

[0141] S20254: Based on the characteristics of agent communication relationships and agent state time sequence characteristics, the agent resource state prediction module of the agent resource state prediction model is used to predict the agent resource state and obtain the dynamic resource state prediction matrix of task capability requirements.

[0142] S20255: Input the dynamic resource state prediction matrix into the scheduling scheme generation model in the agent heterogeneous management model.

[0143] In one optional implementation, based on the dynamic resource state prediction matrix, a scheduling scheme generation model is used to generate an initial scheduling scheme for heterogeneous agents with task capability requirements, and this initial scheme is input into the scheduling scheme optimization model within the agent heterogeneous management model, including:

[0144] S20261: The initialization module for generating the model using the scheduling scheme encodes the scheduling scheme into individual vectors of the IBOA algorithm and sets the IBOA population parameters and the maximum number of iterations.

[0145] S20262: Based on the IBOA population parameters, the initial IBOA population is obtained by initializing using the Tent chaotic mapping sequence; each IBOA individual in the IBOA population corresponds to an alternative scheduling scheme.

[0146] The formula is:

[0147]

[0148] In the formula, For the initial IBOA population, the first i One initial IBOA individual; For the first i One chaotic variable; These are the upper and lower bounds of the search space; i For IBOA individual indicators;

[0149]

[0150] In the formula, For the first i- One chaotic variable; compared with random initialization, chaotic initialization can ensure that the population is evenly distributed in the solution space, thus enhancing diversity.

[0151] S20263: Iterative optimization module for generating models using scheduling schemes, setting the fitness function of the IBOA algorithm;

[0152] The formula is:

[0153]

[0154] In the formula, For IBOA individuals X fitness value; The fitness function; Cost of task delay; Costs associated with resource overload; For communication costs; The fitness weighting coefficient can be adjusted according to actual needs; XFor IBOA individuals, refer to the parameters;

[0155] Task delay cost:

[0156] From the dynamic resource state prediction matrix P Extracting intelligent agents j CPU and memory load forecast sequences within future time windows: P CPU, j , :] and P [MEM, j , :];

[0157] Task () l The resource requirements are "overlaid" onto this prediction sequence, simulating the task. l The impact of adding [something] on future load can be calculated more accurately based on this "overlapping" dynamic load curve. l The projected start and finish times are calculated, taking into account queuing delays caused by other upcoming tasks (which are also included in the prediction matrix).

[0158] That is, all tasks l The sum of (estimated completion time - expected deadline) (can be 0 or negative if completed ahead of schedule); the prediction matrix makes the estimation of delay costs no longer static, but dynamic and takes into account future competition. The algorithm will tend to schedule tasks to agents that are "relatively idle in the future" rather than agents that are "currently idle".

[0159] Resource overload cost:

[0160] For scheduling schemes X Similarly, the impact of adding each task on the future load sequence of each agent is simulated, and all agents are traversed. j And all future time steps k Check the superimposed load CPU, j , k Has the safety threshold been exceeded?

[0161] It is the integral (or summation) of all parts exceeding the threshold, which is more refined than the simple binary penalty of "whether it is overloaded". It not only punishes overload behavior, but also "the severity of overload" and "the duration of overload". The algorithm will actively avoid agents that are not currently busy, but are predicted to reach the bottleneck soon, thereby achieving smooth load balancing and avoiding performance fluctuations.

[0162] Communication cost:

[0163] For scheduling schemes X Tasks that require frequent communication ( l , m They are respectively scheduled to intelligent agents. j and n From the dynamic resource state prediction matrix P In the process, extract the connection agent. j and n Network link bandwidth usage prediction sequence within future time windows: P [Bandwidth, Link( j , n ), :];

[0164] Task pair ( l , m The communication demands of the link are "overlaid" onto the bandwidth prediction sequence of that link. The total communication delay is calculated based on this "overlapped" dynamic bandwidth utilization rate. If the overlapped bandwidth demand exceeds the link capacity, a huge penalty is imposed. The algorithm avoids scheduling high-communication-demand tasks to links that are "about to be squeezed out by other data transmission tasks in the future," and instead looks for paths with "relatively abundant future bandwidth resources," thus ensuring communication quality.

[0165] S20264: Based on the dynamic resource state prediction matrix, the fitness function is used to obtain the fitness value of each initial IBOA individual, and the initial IBOA individual with the best fitness value is taken as the optimal solution.

[0166] S20265: Conduct foraging, mating, reproduction, and milking behaviors to iteratively update the initial IBOA population and obtain an updated IBOA population.

[0167] The formula is:

[0168]

[0169] In the formula, Number of iterations t+ The foraging behavior of 1 was obtained as the first i One updated IBOA individual; This is the convergence factor, which controls the strength of the approach to the optimal solution; Number of iterations t The optimal solution; t This represents the current iteration number; Number of iterations t The i A new IBOA individual is created, with the initial IBOA individual being created during the first iteration update.

[0170]

[0171] In the formula, These are the maximum and minimum values ​​of the convergence factor; This represents the maximum number of iterations. t This represents the current iteration number; , To adjust the parameters; It is the hyperbolic tangent function;

[0172]

[0173] In the formula, Number of iterations t+ The mating behavior of 1 resulted in the first i One updated IBOA individual; The mother and father IBOA individuals are randomly selected, generally the two IBOA individuals with the best fitness. A random number between [0, 1];

[0174]

[0175] In the formula, Number of iterations t+ The reproductive behavior of 1 resulted in the first i One updated IBOA individual;

[0176]

[0177] In the formula, Number of iterations t+ The milking behavior of 1 resulted in the first i One updated IBOA individual; This is the disturbance control coefficient, a small constant (e.g., 0.01), which controls the strength of the disturbance.

[0178]

[0179] In the formula, Number of iterations t+ The dynamic inverse of 1 yields the first i One updated IBOA individual; Number of iterations t The i The reverse solution for each updated IBOA individual; The fitness function;

[0180]

[0181] In the formula, Number of iterationst The reverse center point determined by the dynamic boundary of the current search space; It is a dynamic inverse factor that increases with iteration (intensified inverse exploration in later stages).

[0182] Integrating foraging behavior, mating behavior, reproductive behavior, milking behavior, and dynamic reverse behavior to obtain the first i A new IBOA individual is generated, resulting in a new IBOA population;

[0183] S20266: Based on the dynamic resource state prediction matrix, use the fitness function to obtain the fitness value of each updated IBOA individual, and update the updated IBOA individual with the best fitness value as the optimal solution;

[0184] S20267: When the number of iterations reaches the maximum number of iterations or the fitness value of the optimal solution meets the requirements, terminate the iterative update of the IBOA population and output the optimal solution of the current iteration;

[0185] S20268: Decode the individual vectors of the IBOA individuals corresponding to the optimal solution to obtain the initial scheduling scheme of the heterogeneous agents that generate task capability requirements, and input it into the scheduling scheme optimization model in the heterogeneous agent management model.

[0186] In one alternative implementation, based on the dynamic resource state prediction matrix and the initial scheduling scheme, a scheduling scheme optimization model is used to optimize the initial scheduling scheme, generating a final scheduling scheme for heterogeneous agents with different task capability requirements, including:

[0187] S20271: Input the task capability requirements into the policy network optimization module of the scheduling scheme optimization model, and use the policy network optimization module to adjust the policy network of the scheduling scheme optimization module to obtain an updated policy network.

[0188] S20272: Input the dynamic resource state prediction matrix and the initial scheduling scheme into the state space of the scheduling scheme optimization module;

[0189] S20273: Based on the action space of the scheduling scheme optimization module, use the updated policy network to make scheduling scheme optimization decisions, and obtain scheduling scheme optimization decisions, such as: {Execute scheme A, Execute scheme B, Delay execution for 5 minutes, Request backup resources};

[0190] S20274: Based on the scheduling scheme optimization decision, optimize the initial scheduling scheme to generate the final scheduling scheme of heterogeneous intelligent agents with task capability requirements;

[0191] In this embodiment, the policy network provides the expected long-term reward for each action. This reward considers not only the efficiency of completing the current task, but also the impact of the decision on the future system state (such as whether it will lead to extreme resource shortages in a certain period of time in the future). The updated policy network selects the action with the highest Q value as the final decision. For example, if the Q value of executing plan A is the highest, then plan A is adopted. If the Q value of requesting backup resources is the highest, it indicates that potential risks are foreseen, and the backup plan will be activated first.

[0192] In one optional implementation, empirical data after the execution of the scheduling scheme is collected, and based on this empirical data, the agent heterogeneous management model is self-evolved and trained to obtain an updated agent heterogeneous management model, which then waits to receive the next task information, including:

[0193] S2031: Collect experience data after the execution of the scheduling scheme, store the experience data in the experience replay pool of the scheduling scheme optimization model of the agent heterogeneous management model, and periodically extract a number of experience data from the experience replay pool.

[0194] S2032: Based on some empirical data, perform self-evolution training on the agent heterogeneous management model in the cache area, and extract the model update amount of the updated agent heterogeneous management model.

[0195] S2033: If the model update amount is greater than the model update amount threshold, the agent heterogeneous management model in the memory area of ​​the management server is adjusted, the corresponding updated agent heterogeneous management model is stored, and the system waits to receive the next task information.

[0196] This invention also provides a heterogeneous management device for intelligent agents based on a unified protocol, referring to... Figure 3 The diagram shows a functional unit diagram of a heterogeneous management device 300 for intelligent agents based on a unified protocol according to the present invention. The device may include the following units:

[0197] Initialization unit 301 is used to define a unified protocol framework for agent communication in the enterprise's management server, register several heterogeneous agents of the enterprise, obtain a heterogeneous agent resource library, and build an agent heterogeneous management model.

[0198] The heterogeneous agent management unit 302 is used to generate a scheduling scheme for heterogeneous agents based on a unified protocol framework and a heterogeneous agent resource library, according to the user's task information, using the heterogeneous agent management model, and to execute the scheduling scheme.

[0199] The self-evolutionary training unit 303 is used to collect experience data after the execution of the scheduling scheme, and to perform self-evolutionary training on the agent heterogeneous management model based on the experience data to obtain an updated agent heterogeneous management model, and wait to receive the next task information.

[0200] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.

[0201] Memory, used to store computer programs;

[0202] When a processor executes a program stored in memory, it implements the heterogeneous management method for intelligent agents based on a unified protocol according to the present invention.

[0203] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0204] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0205] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the heterogeneous management method for intelligent agents based on a unified protocol according to embodiments of the present invention.

[0206] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable hardware devices (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0207] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0208] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0209] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0210] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "And / or" indicates that either one or both can be chosen. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0211] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A heterogeneous management method for intelligent agents based on a unified protocol, characterized in that, The method includes: In the enterprise's management server, a unified protocol framework for agent communication is defined, several heterogeneous agents of the enterprise are registered, a heterogeneous agent resource library is obtained, and an agent heterogeneous management model is constructed. The heterogeneous management model for intelligent agents includes an intelligent agent resource state prediction model based on deep learning algorithms, a scheduling scheme generation model based on swarm intelligence optimization algorithms, a scheduling scheme optimization model based on reinforcement learning algorithms, and an intelligent agent anomaly detection model based on deep learning algorithms. The agent resource state prediction model is constructed based on the GNN-LSTM-MLP algorithm, and includes an agent communication relationship feature extraction module constructed based on the GNN algorithm, an agent state temporal feature extraction module constructed based on the LSTM algorithm, and an agent resource state prediction module constructed based on the MLP algorithm. The scheduling scheme generation model is constructed based on the IBOA algorithm, and the scheduling scheme generation model includes an initialization module, an iterative optimization module, and an optimal decoding module. The scheduling scheme optimization model is constructed based on the MPO-PPO algorithm, and includes a policy network optimization module based on the MPO algorithm and a scheduling scheme optimization module based on the PPO algorithm. The scheduling scheme optimization module is equipped with an experience replay pool. The agent anomaly detection model is constructed based on the RF-Attention-SVM algorithm, and includes a key feature extraction module based on the RF algorithm, a weighted fusion module based on the Attention mechanism, and an agent anomaly detection module based on the SVM algorithm. Based on a unified protocol framework and a heterogeneous agent resource library, and according to the user's task information, a heterogeneous agent management model is used to generate a scheduling scheme for heterogeneous agents, and then execute the scheduling scheme, including: It receives task information input by the user, converts the task information into a task request message based on a unified protocol framework, and transmits the task request message to the management server. The system invokes a pre-defined domain knowledge graph to decompose the task request messages within the management server, thereby obtaining at least one task capability requirement. Based on the task capability requirements, query the heterogeneous intelligent agent resource library, filter out all candidate intelligent agents that meet the requirements, and obtain the corresponding candidate intelligent agent pool. Based on a unified protocol framework, the status report messages of each candidate agent in the candidate agent pool are collected, and several status report messages are input into the agent resource status prediction model in the agent heterogeneous management model. Based on several status report messages, the agent resource status prediction model is used to generate a dynamic resource status prediction matrix of task capability requirements, and input into the scheduling scheme generation model in the agent heterogeneous management model. Based on the dynamic resource state prediction matrix, an initial scheduling scheme for heterogeneous agents with task capacity requirements is generated using a scheduling scheme generation model. This initial scheme is then input into the scheduling scheme optimization model within the heterogeneous agent management model, including: The initialization module of the model is generated using a scheduling scheme, which encodes the scheduling scheme into individual vectors of the IBOA algorithm and sets the IBOA population parameters and the maximum number of iterations. Based on the IBOA population parameters, the initial IBOA population is obtained by initializing using the Tent chaotic mapping sequence; each IBOA individual in the IBOA population corresponds to an alternative scheduling scheme. The formula is: In the formula, For the initial IBOA population, the first i One initial IBOA individual; For the first i One chaotic variable; These are the upper and lower bounds of the search space; i For IBOA individual indicators; In the formula, For the first i- One chaotic variable; The iterative optimization module of the model is generated using a scheduling scheme, and the fitness function of the IBOA algorithm is set. The formula is: In the formula, For IBOA individuals X fitness value; The fitness function; Cost of task delay; Costs associated with resource overload; For communication costs; For fitness weighting coefficients; X For IBOA individuals, refer to the parameters; Based on the dynamic resource state prediction matrix, the fitness function is used to obtain the fitness value of each initial IBOA individual, and the initial IBOA individual with the best fitness value is taken as the optimal solution. The initial IBOA population is iteratively updated by conducting foraging, mating, reproduction, and milking behaviors to obtain an updated IBOA population. The formula is: In the formula, Number of iterations t+ The foraging behavior of 1 was obtained as the first i One updated IBOA individual; This is the convergence factor, which controls the strength of the approach to the optimal solution; Number of iterations t The optimal solution; t This represents the current iteration number; Number of iterations t The i A new IBOA individual is created, with the initial IBOA individual being created during the first iteration update. In the formula, These are the maximum and minimum values ​​of the convergence factor; This represents the maximum number of iterations. t This represents the current iteration number; , To adjust the parameters; It is the hyperbolic tangent function; In the formula, Number of iterations t+ The mating behavior of 1 resulted in the first i One updated IBOA individual; The mother and father IBOA individuals are randomly selected, generally the two IBOA individuals with the best fitness. A random number between [0, 1]; In the formula, Number of iterations t+ The reproductive behavior of 1 resulted in the first i One updated IBOA individual; In the formula, Number of iterations t+ The milking behavior of 1 resulted in the first i One updated IBOA individual; These are the disturbance control coefficients; In the formula, Number of iterations t+ The dynamic inverse of 1 yields the first i One updated IBOA individual; Number of iterations t The i The reverse solution for each updated IBOA individual; The fitness function; In the formula, Number of iterations t The reverse center point determined by the dynamic boundary of the current search space; It is a dynamic inverse factor; Integrating foraging behavior, mating behavior, reproductive behavior, milking behavior, and dynamic reverse behavior to obtain the first i A new IBOA individual is generated, resulting in a new IBOA population. Based on the dynamic resource state prediction matrix, the fitness function is used to obtain the fitness value of each updated IBOA individual, and the updated IBOA individual with the best fitness value is updated as the optimal solution. When the number of iterations reaches the maximum number of iterations or the fitness value of the optimal solution meets the requirements, the iterative update of the IBOA population is terminated, and the optimal solution of the current iteration is output. Decode the individual vectors of the IBOA individuals corresponding to the optimal solution to obtain the initial scheduling scheme of the heterogeneous agents that generate task capability requirements, and input it into the scheduling scheme optimization model in the agent heterogeneous management model. Based on the dynamic resource state prediction matrix and the initial scheduling scheme, the scheduling scheme optimization model is used to optimize the initial scheduling scheme and generate the final scheduling scheme of heterogeneous agents with task capability requirements. Based on a unified protocol framework and according to the final scheduling scheme, task instruction messages for the target intelligent agents that require task capabilities are generated, and the task instruction messages are sent to the corresponding target intelligent agents. Using the target intelligent agent, receive task instruction messages, and send business data messages to other target intelligent agents with the same task capability requirements based on a unified protocol framework; Based on business data messages, all target agents with the same task capability requirements synchronously execute the task instruction messages of the final scheduling scheme, and during execution, each target agent sends the latest status report message to the management server. Based on the latest status report, the agent anomaly detection module is used to generate the agent anomaly detection result for the target agent. If the agent anomaly detection result indicates that an anomaly exists, rescheduling is triggered, and the process returns to the agent resource status prediction step. Collect experience data after the execution of the scheduling scheme, and perform self-evolution training on the agent heterogeneous management model based on the experience data to obtain an updated agent heterogeneous management model, and wait to receive the next task information.

2. The heterogeneous management method for intelligent agents based on a unified protocol according to claim 1, characterized in that, In the enterprise's management server, a unified protocol framework for agent communication is defined, a heterogeneous agent management model is constructed, and several heterogeneous agents within the enterprise are registered to obtain a heterogeneous agent resource library, including: In the enterprise's management server, a structured message format, communication interface, interaction process, and protocol stack are defined to obtain a unified protocol framework for intelligent agent communication. Based on a unified protocol framework, dynamic registration and capability modeling are performed on several heterogeneous intelligent agents in the management server to obtain a heterogeneous intelligent agent resource library, and a heartbeat mechanism is set for each intelligent agent. Based on a unified protocol framework, an intelligent agent heterogeneous management model is constructed using artificial intelligence algorithms, and the intelligent agent heterogeneous management model is connected to a heterogeneous intelligent agent resource library.

3. The heterogeneous management method for intelligent agents based on a unified protocol according to claim 2, characterized in that, The message format includes a message header, a message body, and metadata; The message header includes a message ID, a source agent ID, a target agent, a timestamp, and a message type; The message body includes the actual data for the current message type; The metadata includes the QoS requirements and security credentials for the current message body.

4. The heterogeneous management method for intelligent agents based on a unified protocol according to claim 3, characterized in that, The message types include registration message type, heartbeat message type, task request message type, task instruction message type, status report message type, and business data message type.

5. The heterogeneous management method for intelligent agents based on a unified protocol according to claim 4, characterized in that, Based on a unified protocol framework, dynamic registration and capability modeling are performed on several heterogeneous agents in the management server to obtain a heterogeneous agent resource library. A heartbeat mechanism is then set for each agent, including: Based on a unified protocol framework, the system collects registration messages from several heterogeneous pre-set and / or new intelligent agents in the management server, and performs security verification on the identity credentials of the intelligent agents based on the registration messages. If the security verification of the current agent passes, proceed to the next step; otherwise, delete the corresponding agent, issue an alarm signal, and terminate the registration of the corresponding agent. Based on the registration message, the corresponding capability list is parsed, the capability list is converted into the corresponding intelligent agent capability model, and the intelligent agent capability model is stored in the heterogeneous intelligent agent resource library. By traversing all agents, a heterogeneous agent resource library including the capability models of all agents is obtained, and a heartbeat mechanism is set for each agent, sending heartbeat messages to the management server based on a unified protocol framework.

6. The heterogeneous management method for intelligent agents based on a unified protocol according to claim 5, characterized in that, Collect empirical data after the execution of the scheduling scheme, and based on this data, perform self-evolutionary training on the agent heterogeneous management model to obtain an updated agent heterogeneous management model, and wait to receive the next task information, including: Collect experience data after the execution of the scheduling scheme, store the experience data in the experience replay pool of the scheduling scheme optimization model of the agent heterogeneous management model, and periodically extract a number of experience data from the experience replay pool. Based on some empirical data, the agent heterogeneous management model is trained in a cache area through self-evolution, and the model update amount of the updated agent heterogeneous management model is extracted. If the model update amount exceeds the model update amount threshold, the agent heterogeneous management model in the memory area of ​​the management server will be adjusted, the corresponding updated agent heterogeneous management model will be stored, and the server will wait to receive the next task information.

7. A heterogeneous management device for intelligent agents based on a unified protocol, used to implement the heterogeneous management method for intelligent agents as described in any one of claims 1-6, characterized in that, The device includes: The initialization unit is used to define a unified protocol framework for agent communication in the enterprise's management server, register several heterogeneous agents of the enterprise, obtain a heterogeneous agent resource library, and build a heterogeneous agent management model. The agent heterogeneous management unit is used to generate a scheduling scheme for heterogeneous agents based on a unified protocol framework and a heterogeneous agent resource library, according to the user's task information, using the agent heterogeneous management model, and then execute the scheduling scheme. The self-evolutionary training unit is used to collect experience data after the execution of the scheduling scheme, and to perform self-evolutionary training on the agent heterogeneous management model based on the experience data to obtain an updated agent heterogeneous management model, and wait to receive the next task information.

Images