Operation device and operation method of multi-agent system, and storage medium

Abstract

The invention provides a running device and running method of a multi-agent system and a storage medium, and the running device comprises a distributed storage system which comprises a storage server used for storing flow state information of the multi-agent system according to a distributed consistency storage protocol; the flow state information comprises agent information and scene information in the environment; the agent calculation system comprises an operation server and is used for calculating an operation task of at least one agent in the multi-agent system based on the flow state information in the storage server; the management agent calculation service system is used for calculating the operation task of the environment of the multi-agent system based on the flow state information in the storage server to obtain an environment operation result; and all the storage servers are also used for storing the environment running result according to the distributed consistency storage protocol. According to the invention, the operation server can be added according to the computing power demand of the intelligent agent to overcome the computing power bottleneck.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an operating device, an operating method, and a computer-readable storage medium for a multi-agent system. Background Art

[0002] Multi-agent technology has been a key research frontier in the development of artificial intelligence (AI). Research on multi-agent technology focuses on the consistent characteristics and evolutionary patterns exhibited by the multi-agent system as a whole, based on the interactions between the agents and the environment, as well as between different agents, when multiple agents exist in an environment (or world). These behaviors, such as competition, cooperation, deception, alliances, and betrayal, are captured by AI models capable of inputting and outputting information.

[0003] Existing technologies for multi-agent systems fall into two main categories: one is multi-agent simulation, which simplifies and abstracts individual agents; the other is the use of reinforcement learning to synchronously train multiple agents. These technologies require the specialized construction of an environment for the agents to run, which carries significant computational complexity. As the number of agents and the complexity of their behaviors increase, the computational burden on the environment also increases, ultimately becoming a performance bottleneck for the entire multi-agent system. Furthermore, these technologies are all based on computing interactive feedback based on agent behaviors within a collection timeframe. The interactions between different agents and between agents and the environment are asynchronous, significantly different from the real world. Furthermore, once the environment (or world) design in these technologies is determined, its content and logic are generally immutable, making it difficult to support dynamic additions of new agents and new action capabilities during operation. Summary of the Invention

[0004] Embodiments of the present invention provide an operating device, an operating method, and a computer-readable storage medium for a multi-agent system to solve at least one problem existing in the prior art.

[0005] In order to achieve the above object, the present invention adopts the following scheme:

[0006] According to a first aspect of an embodiment of the present invention, there is provided an operating device for a multi-agent system, comprising:

[0007] A distributed storage system comprising at least one storage server; all of the storage servers are configured to store flow information of a multi-agent system according to a distributed consistency storage protocol; the flow information includes agent information and scene information in an environment;

[0008] An agent computing system comprising at least one computing server; at least one computing server being configured to compute an operating task of at least one agent in the multi-agent system based on flow information in the storage server;

[0009] A management agent computing service system is used to calculate the operating tasks of the environment of the multi-agent system based on the flow information in the storage server to obtain the environment operating results; all the storage servers are also used to store the environment operating results according to the distributed consistency storage protocol.

[0010] In some embodiments, the operating device of the multi-agent system is characterized in that it further includes a management server;

[0011] The storage server is further configured to receive an instruction for an agent to enter the multi-agent system, and the management server is further configured to allocate a computing server to perform calculations on the running tasks of the entering agent; and / or

[0012] The storage server is further configured to receive an instruction for an agent to exit the multi-agent system, and the management server is further configured to stop the corresponding computing server from computing the running tasks of the exiting agent; and / or

[0013] The management server is also used to adjust the computing server in the intelligent computing system to the storage server in the distributed storage system, or to adjust the storage server in the distributed storage system to the computing server in the intelligent computing system according to the computing power of the intelligent computing system and the storage situation of the distributed storage system.

[0014] In some embodiments, the management agent computing service system is also used to store the environmental evolution law information of the multi-agent system, and to calculate the operating tasks of the environment of the multi-agent system based on the flow information in the storage server and the environmental evolution law information.

[0015] In some embodiments, the management agent computing service system includes at least one management agent computing service module, and the management agent computing service system is a distributed computing system; and / or,

[0016] The intelligent agent computing system is a distributed computing system.

[0017] In some embodiments, the management agent computing service system is a distributed computing system, and the management agent computing service system includes at least one management agent computing service module, each of the management agent computing service modules is a containerized computing service module; and / or, each of the storage servers and each of the computing servers are physical servers.

[0018] In some embodiments, the running task of the agent includes at least one of a computing task of the agent's action intention to the environment and a computing task of the agent's mental modeling of other agents; and / or,

[0019] The running tasks of the environment include at least one of the following: computing tasks for changes in the state of the environment caused by the actions of the intelligent agent, computing tasks for natural changes in the environment according to information on the laws of environmental evolution, and computing tasks for determining the results of the joint actions of multiple intelligent agents; and / or,

[0020] The agent information includes the agent's state information, the agent's action intention information, and the agent's action result information; the scene information in the environment includes the state information of objects in the scene in the environment; and / or,

[0021] Each of the intelligent agents includes a predetermined artificial intelligence agent model and an intelligent agent data structure.

[0022] In some embodiments, the agents in the multi-agent system communicate asynchronously with the corresponding computing servers through subscription / publishing; and / or,

[0023] The management agent computing service system and the corresponding operation server realize asynchronous communication through subscription / publishing.

[0024] According to a second aspect of an embodiment of the present invention, a method for operating a multi-agent system is provided, which is applicable to the operating device of the multi-agent system according to any embodiment of the present invention, and the operating method includes:

[0025] The first agent in the multi-agent system reads the flow state information of the multi-agent system stored in the storage server to obtain first flow state information;

[0026] The first agent calculates the running task of the first agent using the computing server and according to the first flow state information to obtain an interaction request;

[0027] Storing the information of the interaction request in a storage server according to a distributed consistency storage protocol;

[0028] Acquire the interaction request information from the storage server, and read the flow state information of the multi-agent system stored in the storage server according to the interaction request information to obtain second flow state information;

[0029] Utilizing a management agent computing service system and calculating an operating task of an environment in the multi-agent system according to the second flow state information and the acquired information of the interaction request, to obtain an environment operating result;

[0030] Storing the environment operation result in a storage server according to the distributed consistency storage protocol;

[0031] The first agent reads information corresponding to the environment operation result from the storage server;

[0032] The first agent obtains the state change information of the first agent by using the computing server and information corresponding to the environment operation result;

[0033] The state change information of the first agent is stored in a storage server according to a distributed consistency storage protocol.

[0034] In some embodiments, the management agent computing service system is used to calculate the operating tasks of the environment in the multi-agent system based on the second flow information and the information of the interaction request obtained to obtain the environment operating results, including: using the management agent computing service system to calculate the operating tasks of the environment in the multi-agent system based on the second flow information, the information of the interaction request obtained and the environment evolution law information locally stored in the management agent computing service system to obtain the environment operating results.

[0035] In some embodiments, the management agent computing service system is used to calculate the operating tasks of the environment in the multi-agent system based on the second flow information, the information of the interaction request obtained, and the environmental evolution law information stored locally in the management agent computing service system to obtain the environmental operation results, including: using multiple management agent computing service modules in the management agent computing service system and performing distributed computing on the operating tasks of the environment in the multi-agent system based on the second flow information, the information of the interaction request obtained, and the environmental evolution law information stored locally in the management agent computing service system to obtain the environmental operation results.

[0036] In some embodiments, the first intelligent agent uses a computing server and calculates the running task of the first intelligent agent according to the first flow information to obtain an interaction request, including: the first intelligent agent uses multiple computing servers in the intelligent agent computing system and performs distributed computing on the running task of the first intelligent agent according to the first flow information to obtain an interaction request; the first intelligent agent uses the computing server and obtains the state change information of the first intelligent agent according to the information corresponding to the environmental running result, including: the first intelligent agent uses multiple computing servers in the intelligent agent computing system and performs distributed computing on the running task of the first intelligent agent according to the information corresponding to the environmental running result to obtain the state change information of the first intelligent agent.

[0037] In some embodiments, the method for operating a multi-agent system further includes:

[0038] Increase or decrease the number of computing servers in the agent computing system according to the computing power of the agent computing system; and / or,

[0039] Increase or decrease the number of storage servers in the distributed storage system according to the storage situation of the distributed storage system; and / or,

[0040] According to the computing power of the intelligent computing system and the storage situation of the distributed storage system, some of the computing servers in the intelligent computing system are adjusted to storage servers of the distributed storage system, or, some of the storage servers in the distributed storage system are adjusted to computing servers of the intelligent computing system.

[0041] In some embodiments, the method for operating the multi-agent system further includes:

[0042] The first agent in the multi-agent system reads the flow state information of the multi-agent system stored in the storage server to obtain first flow state information, including: the first agent in the multi-agent system queries the flow state information of the multi-agent system stored in the storage server to obtain current information of a first object in the environment of the multi-agent system; the first flow state information includes the current information of the first object;

[0043] The first agent calculates the first agent's operation task using the computing server and based on the first flow information to obtain an interaction request, including: the first agent calculates the first agent's operation task using the computing server and based on the current information of the first object to obtain first action intention information of the first agent with respect to the first object; the interaction request includes the first action intention information;

[0044] Storing the information of the interaction request in a storage server according to a distributed consistency storage protocol includes: storing the first action intention information in a storage server according to a distributed consistency storage protocol;

[0045] Obtaining the interaction request information from the storage server, and reading the flow information of the multi-agent system stored in the storage server according to the interaction request information to obtain second flow information, including: obtaining the first action intention information in the interaction request information from the storage server, and reading the flow information of the multi-agent system stored in the storage server according to the first action intention information to obtain second flow information including relevant information of the first object;

[0046] Utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the second flow state information and the obtained information of the interaction request to obtain an environment operating result, including: utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the first action intention information and relevant information of the first object to obtain an environment operating result including a change result of the first object in the environment;

[0047] Storing the environment operation result in a storage server according to the distributed consistency storage protocol, including: storing information corresponding to the change result in the storage server according to the distributed consistency storage protocol;

[0048] The first agent reads information corresponding to the environment operation result from the storage server, including: the first agent reads information of a change result of the first object in the environment from the storage server; the information corresponding to the environment operation result includes information of a change result of the first object in the environment;

[0049] The first agent obtains the state change information of the first agent using the computing server and information corresponding to the result of the operation of the environment, including: the first agent determines the state change information of the first agent using the computing server and information corresponding to the result of the change of the first object in the environment;

[0050] Storing the state change information of the first agent in a storage server according to a distributed consistency storage protocol includes: storing the change information of the first category state of the first agent in a storage server according to a distributed consistency storage protocol.

[0051] In some embodiments, the management agent computing service system is used to calculate the operating tasks of the environment in the multi-agent system based on the first action intention information and the relevant information of the first object to obtain the environment operating results including the change results of the first object in the environment, including: obtaining the corresponding environment evolution law information locally from the management agent computing service system based on the first action intention information; using the management agent computing service system to calculate the operating tasks of the environment in the multi-agent system based on the first action intention information, the relevant information of the first object, and the environment evolution law information to obtain the change results of the first object in the environment; using the management agent computing service system to calculate based on the change results of the first object in the environment to obtain the state change information of the first agent; the environment operating results include the change results of the first object and the change information of the first category state of the first agent;

[0052] storing the environment operation result in a storage server according to the distributed consistency storage protocol, further comprising: storing change information of the first category state of the first agent in the storage server according to the distributed consistency storage protocol;

[0053] The first agent reads information corresponding to the environment operation result from the storage server, further comprising: the first agent reads change information of the first category state of the first agent from the storage server;

[0054] The first agent determines the state change information of the first agent using the computing server and information about the change result of the first object in the environment, including: the first agent determines the change information of the second category state of the first agent using the computing server and information about the change result of the first object in the environment and the change information of the first category state of the first agent;

[0055] Storing the state change information of the first agent in the storage server according to the distributed consistency storage protocol also includes: storing the change information of the second category state of the first agent in the storage server according to the distributed consistency storage protocol.

[0056] In some embodiments, the operation method also includes: the first agent in the multi-agent system reads the external state information of the second agent in the multi-agent system stored in the storage server; the first agent uses the computing server and calculates the mental modeling task in the operation task of the first agent based on the external state information of the second agent to obtain an interaction request about mental information; and stores the interaction request about mental information to the storage server according to the distributed consistency storage protocol.

[0057] In some embodiments, the operation method further includes: the second agent in the multi-agent system reads the flow information of the multi-agent system stored in the storage server to obtain third flow information; the second agent uses the computing server and calculates the operation task of the second agent based on the third flow information to obtain the interaction request of the second agent; stores the interaction request information of the second agent in the storage server according to the distributed consistency storage protocol; obtains the interaction request information of the second agent from the storage server, and reads the flow information of the multi-agent system stored in the storage server according to the interaction request information of the second agent to obtain fourth flow information;

[0058] Utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the second flow state information and the obtained information of the interaction request to obtain an environment operating result, including: utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the second flow state information, the fourth flow state information, the information of the interaction request of the first agent, and the information of the interaction request of the second agent to obtain an environment operating result;

[0059] The operation method also includes: the second intelligent agent reads the information corresponding to the environment operation result from the storage server; the second intelligent agent uses the computing server and the information corresponding to the environment operation result to obtain the state change information of the second intelligent agent; and stores the state change information of the second intelligent agent to the storage server according to the distributed consistency storage protocol.

[0060] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the steps of the method described in any embodiment of the present invention are implemented.

[0061] The multi-agent system operating device, operating method, and computer-readable storage medium of the present invention store flow information based on a distributed consistency storage protocol. This flow information can be read when computing both the agent's and the environment's running tasks. The entry and exit of agents do not affect the stored flow information. By using computing servers to compute the agent's running tasks, computing servers can be flexibly added or reduced based on the agent's computing power requirements, avoiding bottlenecks caused by the increasing number and complexity of agents. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0063] Figure 1 is a schematic structural diagram of an operating device of a multi-agent system according to an embodiment of the present invention;

[0064] Figure 2 is a flow chart of an operating method of a multi-agent system according to an embodiment of the present invention;

[0065] Figure 3 1 is a schematic diagram showing the structural relationship of an operating device of a multi-agent system according to an embodiment of the present invention;

[0066] Figure 4 This is a schematic diagram of the physical implementation topology connection of a multi-agent system in one embodiment of the present invention. DETAILED DESCRIPTION

[0067] To make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0068] Existing technologies for multi-agent systems primarily fall into two categories. One involves simplified, abstract multi-agent simulations of individuals. These systems primarily target scenarios with a large number of agents, focusing more on overall statistical characteristics and the emergence of group behavior. The modeling and interaction design of the agents themselves are relatively simple. The other involves synchronized multi-agent training schemes similar to reinforcement learning. These systems require a specialized understanding of the world's content and operating rules, while designing the same or different reinforcement learning models for each individual. Joint training is performed through reinforcement learning to optimize the individual strategies or reward functions.

[0069] A common feature of both of these approaches is the need for specialized environments to run the agents, which also carries significant computational complexity. As the number of agents and the complexity of their behaviors increase, the environment significantly increases the computational burden, becoming a performance bottleneck for the entire system. Another issue is that in the real world, interactions between agents and their environments often occur asynchronously. Both approaches typically rely on collecting the behaviors of all agents over a period of time and then calculating and distributing feedback in a unified manner, an approach that differs from the real world. A more significant drawback is that once the environment / world design is finalized, its content and logic are largely immutable. Therefore, current multi-agent systems struggle to support requirements such as dynamically adding new agents or capabilities during runtime.

[0070] To this end, the present invention provides an operating device for a multi-agent system to solve at least one of the above problems. Figure 1 This is a schematic diagram of the structure of the operating device of the multi-agent system according to an embodiment of the present invention. Figure 1 , the device may include: a distributed storage system 110, a management intelligent agent computing service system 120 and an intelligent agent computing system 130.

[0071] The distributed storage system 110 includes at least one storage server 111; all the storage servers 111 are used to store the flow information of the multi-agent system according to the distributed consistency storage protocol; the flow information includes the agent information and the scene information in the environment.

[0072] The intelligent agent computing system 130 includes at least one operation server 131; at least one of the operation servers 131 is used to calculate the operating tasks of at least one intelligent agent in the multi-agent system based on the flow information in the storage server 111. Based on this, the operation server 131 can be dynamically increased or decreased according to the computing power required by the intelligent agents in the multi-agent system. The flow information includes intelligent agent information and scene information in the environment. Among them, the operating tasks of the intelligent agent may include the calculation task of the intelligent agent's action intention to the environment, the calculation task of the intelligent agent's mental modeling of other intelligent agents, etc. The action intention can be the interaction between the intelligent agent and the world (or environment), for example, the action intention of the intelligent agent to eat an apple. Mental modeling can be the process of an intelligent agent establishing cognition based on its understanding of the behavior of other intelligent agents, such as the establishment of cognition of one intelligent agent on whether another intelligent agent does something.

[0073] In a further embodiment, a management server can be used to adjust the number of computing servers 131 in the intelligent computing system 130. The management server can adjust the number of computing servers 131 based on the computing power of the intelligent computing system 130. For example, when the tasks of the intelligent agents are more complex or the number of intelligent agents is large, more computing power is required, and new computing servers 131 can be added. In a further embodiment, the number of storage servers 111 in the distributed storage system 110 can also be adjusted by the management server. The management server can adjust the number of storage servers 111 based on storage requirements. In other embodiments, the type of server can be adjusted, such as changing the computing server 131 to the storage server 111. For example, when the intelligent agents have fewer actions and more states, less computing power and more storage space are required. In this case, the number of computing servers 131 can be reduced and the number of storage servers 111 can be increased.

[0074] The management agent computing service system 120 is configured to calculate the operating tasks of the multi-agent system environment based on the flow information in the storage server 111 to obtain the environment operating results; all the storage servers 111 are further configured to store the environment operating results according to the distributed consistency storage protocol. The agent computing system can be a distributed computing system, and multiple storage servers 111 can perform distributed computing on the agent operating tasks. When an agent intends to perform an action on the environment, the environment operating task can include the environment's feedback on the agent's action. For example, if a virtual human agent kicks a ball and the ball rolls, the environment operating task needs to calculate the ball's rolling, collision, etc. In addition to flow information, the information based on which the environment operating tasks are calculated can also include other required information, such as world laws (such as gravity, fluid flow, and object breakage laws). World laws can also be stored in the management agent computing service system 120, but it is not ruled out that they can be stored in other locations. In addition, when multiple agents act together on the environment, the environment's operational tasks may include making decisions about the results of the actions of the multiple agents. For example, in a nameplate-tearing game, the management agent computing service system 120 may calculate the result of whether the nameplate can be torn off. The management agent computing service system 120 can run a management agent, which can be used to provide a program algorithm for calculating the results of the environment's operation. The management agent's calculation content includes but is not limited to changes in the environmental state caused by the agent's actions, natural changes in the environment that occur according to evolutionary laws, and the result judgment when multiple agents act together. The interaction of the agent's actions is similar to sending a decision intention to the "management agent", and the "management agent" calculates the result. For example, the agent pushes an object with the intention of "wanting to push the object", and the management agent calculates the result that "the object is too heavy, and the agent itself is pushed back and slides."

[0075] The intelligent agents of each embodiment can be virtual humans or non-virtual humans. For non-virtual humans, action intentions can be converted through the interface, and their actual effects can be generated through the corresponding output channels. Virtual humans can directly interact with the storage server through intention data and scene state data. The difference between non-virtual humans (such as humans, robots, and devices) and virtual humans mainly lies in the addition of intention and perception, which are the conversion from analog to digital and from digital to analog, respectively.

[0076] In this embodiment, flow state information is stored based on a distributed consistency storage protocol. This flow state information can be read from the distributed consistency storage protocol when computing both the agent's running tasks and the environment's running tasks. The entry and exit of agents does not affect the stored flow state information. By utilizing the computing servers in the agent computing system to compute the agent's running tasks, computing servers can be flexibly added or reduced, avoiding bottlenecks caused by the increasing number and complexity of agents. By utilizing the management agent computing service system to compute the environment's running tasks, the environment's running tasks are separated from the calculation of the agent's running tasks, facilitating the adjustment of the agent's computing power.

[0077] In some embodiments, the operating device of the multi-agent system further includes a management server. The storage server can also be used to receive instructions for an agent to enter the multi-agent system, and the management server can also be used to assign a computing server to calculate the operating tasks of the entering agent. In this way, the agent's entry request can be supported. In other embodiments, the storage server can also be used to receive instructions for an agent to exit the multi-agent system, and the management server can also be used to stop the corresponding computing server from calculating the operating tasks of the exiting agent. In this way, the agent's exit request can be supported. In still other embodiments, the management server can also be used to adjust the computing server 131 in the agent computing system to the storage server 111 in the distributed storage system 110, or to adjust the storage server 111 in the distributed storage system 110 to the computing server 131 in the agent computing system 130, based on the computing power of the agent computing system 130 and the storage capacity of the distributed storage system 110.

[0078] The storage of environmental evolution law information can be optimized. In some embodiments, the management agent computing service system can also be used to store the environmental evolution law information of the multi-agent system, and to calculate the operating tasks of the environment of the multi-agent system based on the flow state information and the environmental evolution law information in the storage server. In this embodiment, the environmental evolution law (world law) is stored separately from the flow state information, which is conducive to better operation of the multi-agent system. In other embodiments, it is not ruled out that the world law can be stored in the storage server.

[0079] The specific structure of the management agent computing service system can be optimized. In some embodiments, the management agent computing service system includes at least one management agent computing service module, and the management agent computing service system is a distributed computing system. All management agent computing service modules can be used to perform distributed computing on the operating tasks of the multi-agent system environment to obtain the operating results of the environment. Through distributed computing, the management agent (world simulation) achieves better computing performance.

[0080] Furthermore, in some embodiments, the management agent computing service module can be a containerized computing service module, for example, containerized computing can be performed on a cloud server. The management agent computing service system is a distributed computing system, and the management agent computing service system includes at least one management agent computing service module, and each management agent computing service module is a containerized computing service module. In some other embodiments, each of the storage servers and each of the computing servers is a physical server. In addition, the management server can also be a physical server, and each physical server can be replaced by a cloud server.

[0081] The agent's operational tasks and the environment's operational tasks can be determined based on the agent's actions. In some embodiments, the agent's operational tasks may include at least one of computing the agent's action intentions with respect to the environment and computing the agent's mental modeling of other agents. For example, the agent's action intentions with respect to the environment may be the agent's intention to eat an apple, and the agent's mental modeling of other agents may be one agent's belief in whether other agents will pick up an object. In other embodiments, the environment's operational tasks may include at least one of computing the change in the environment's state caused by the agent's actions, computing the natural changes in the environment according to information about environmental evolution laws, and computing the results of the combined actions of multiple agents. For example, the change in state caused by the agent's intention to eat an apple may be the disappearance of an apple in the environment, and the natural change in the environment according to information about environmental evolution laws may be the change in position caused by the movement of a sphere. In yet other embodiments, the agent information may include the agent's state information, the agent's action intention information (action request information), and the agent's action result information. The scene information within the environment may include the state information of objects within the scene within the environment. Furthermore, intelligent agents are driven by artificial intelligence models. In some embodiments, each intelligent agent may include a predetermined artificial intelligence agent model and agent data structure, such as an AOG model (AND / OR graph model) or a large language model. The AOG model can be a graph structure based on probability and causal relationships, integrating procedural knowledge and objective prior knowledge. When executing a specific node task, the AOG model can process information input (e.g., the results of processing state content in the corresponding world server / storage server), generate a sequence of state change requests, and other processing flows.

[0082] When an agent wishes to utilize a computing server for computation, it can establish a communication connection through a predetermined method. In some embodiments, agents in the multi-agent system can achieve asynchronous communication with the corresponding computing server through a subscription / publish method. In other embodiments, asynchronous communication can be achieved between the management agent computing service system and the corresponding computing server through a subscription / publish method. This subscription / publish method can be a conventional asynchronous communication model. Agents and / or management agents can act as message subscribers or publishers. Through specific topics, publishers and subscribers can be decoupled through subscription services, achieving timely message delivery and asynchronous concurrent processing while avoiding the waste of computing power and communication bandwidth in traditional polling methods. Messages published by publishers can be tagged with specific topics. Relevant messages are placed in a specific queue on the server before being consumed (read). Different subscribers first register with the publisher for several topics to be monitored. Different subscribers will obtain messages from the server according to their needs, or receive push notifications from the server when subscription messages arrive. When all subscribers of a message have read it or the message times out after a long period of inactivity, the message can be deleted from the server.

[0083] In addition, an embodiment of the present invention further provides a method for operating a multi-agent system, which is applicable to the operating device of the multi-agent system described in any of the above embodiments, see Figure 2 The operation method may include the following steps S210 to S290. The method has similarities with the aforementioned operation device. For specific implementation, please refer to the aforementioned operation device embodiment.

[0084] Step S210: The first agent in the multi-agent system reads the flow state information of the multi-agent system stored in the storage server to obtain first flow state information;

[0085] Step S220: The first agent uses the computing server to calculate the running task of the first agent according to the first flow state information to obtain an interaction request;

[0086] Step S230: storing the information of the interaction request in a storage server according to a distributed consistency storage protocol;

[0087] Step S240: Acquire the interaction request information from the storage server, and read the flow state information of the multi-agent system stored in the storage server according to the interaction request information to obtain second flow state information;

[0088] Step S250: using the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the second flow state information and the acquired information of the interaction request, to obtain an environment operating result;

[0089] Step S260: storing the environment operation result in a storage server according to the distributed consistency storage protocol;

[0090] Step S270: the first agent reads information corresponding to the environment operation result from the storage server;

[0091] Step S280: The first agent obtains the state change information of the first agent using the computing server and information corresponding to the environment operation result;

[0092] Step S290: Store the state change information of the first agent to the storage server according to the distributed consistency storage protocol.

[0093] For example, the first agent is a virtual human agent. To complete the task of eating an apple, the first agent queries the world server (storage server) to obtain the apples in the current environment and their specific locations and attributes. The first agent can then send an action intention, such as "eat apple," to the world server. The management agent can receive notification of this action intention through message subscription, calculate the resulting state (the environment's operation results), and perform data operations on the world server, deleting the "apple" object and correspondingly changing the agent's "hunger" level. After the first agent asynchronously queries the world server and discovers that the apple has disappeared and its own hunger level has changed, it can then synchronously update its own cognitive state stored in the "world server."

[0094] Furthermore, the aforementioned step S250, i.e., utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system based on the second flow information and the information obtained about the interaction request to obtain the environment operating result, may specifically include the following step: S251, utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system based on the second flow information, the information obtained about the interaction request, and the environment evolution law information stored locally in the management agent computing service system to obtain the environment operating result. In this embodiment, the environment evolution law information is stored locally in the management agent computing service system, thereby improving the convenience for the management agent to obtain the environment evolution law information.

[0095] Furthermore, the above-mentioned step S251, i.e., utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system based on the second flow information, the information obtained about the interaction request, and the environmental evolution law information stored locally in the management agent computing service system to obtain the environmental operation result, may more specifically include the step: S2511, utilizing multiple management agent computing service modules in the management agent computing service system and performing distributed computing on the operating tasks of the environment in the multi-agent system based on the second flow information, the information obtained about the interaction request, and the environmental evolution law information stored locally in the management agent computing service system to obtain the environmental operation result. In this embodiment, the management agent obtains the result content through distributed computing.

[0096] In a specific embodiment, the above-mentioned step S220, i.e., the first agent uses the computing server and calculates the running task of the first agent according to the first flow information to obtain the interaction request, may include the step: S221, the first agent uses multiple computing servers in the agent computing system and performs distributed computing on the running task of the first agent according to the first flow information to obtain the interaction request. The above-mentioned step S280, i.e., the first agent uses the computing server and obtains the state change information of the first agent according to the information corresponding to the environment operation result, may include the step: the first agent uses multiple computing servers in the agent computing system and performs distributed computing on the running task of the first agent according to the information corresponding to the environment operation result to obtain the state change information of the first agent.

[0097] Furthermore, the management agent computing service module can be a containerized computing service module, which can perform containerized computing on a cloud server. In some embodiments, each of the storage servers, each of the computing servers, and the management server can be a physical server, and cloud servers are not excluded.

[0098] Computing servers and storage servers can be dynamically adjusted. For example, the operating method of a multi-agent system further includes the step of increasing or decreasing the number of computing servers in the agent computing system based on the computing power of the agent computing system. More specifically, the management server can be used to increase or decrease computing servers based on the computing status of the computing servers. For another example, the operating method of a multi-agent system further includes the step of increasing or decreasing the number of storage servers in the distributed storage system based on the storage status of the distributed storage system. More specifically, the management server can be used to increase or decrease storage servers based on the storage status of the storage servers. For another example, the operating method of a multi-agent system further includes the step of adjusting some computing servers in the agent computing system to storage servers of the distributed storage system, or adjusting some storage servers in the distributed storage system to computing servers of the agent computing system, based on the computing power of the agent computing system and the storage status of the distributed storage system. More specifically, the management server can be used to adjust some computing servers to storage servers, or adjusting some storage servers to computing servers, based on the computing status of the computing servers and the storage status of the storage servers.

[0099] In some specific embodiments, the agent changes the state of objects in the environment, such as the agent eats an apple. Figure 2 The specific implementation of each step in the operation method of the multi-agent system shown can be described as follows.

[0100] Step S210, i.e., the first agent in the multi-agent system reads the multi-agent system's flow state information stored in the storage server to obtain first flow state information, may specifically include the following step: S211, where the first agent in the multi-agent system queries the multi-agent system's flow state information stored in the storage server to obtain current information about a first object in the multi-agent system's environment; the first flow state information includes the current information about the first object. For example, for a task in which the agent eats an apple, the current information about the first object may be the apple in the environment, its location, and its status.

[0101] Step S220, i.e., the first agent using the computing server to calculate the first agent's running task based on the first flow information to obtain an interaction request, may specifically include step S221, where the first agent uses the computing server to calculate the first agent's running task based on the first object's current information to obtain the first agent's first action intention information regarding the first object; the interaction request includes the first action intention information. For example, for a task in which the agent is to eat an apple, the first action intention information may be the first agent's desire to eat a particular apple in the environment.

[0102] The above-mentioned step S230, i.e., storing the information of the interaction request to the storage server according to the distributed consistency storage protocol, may specifically include the step: S231, storing the first action intention information to the storage server according to the distributed consistency storage protocol.

[0103] The above-mentioned step S240, i.e., obtaining the information of the interaction request from the storage server, and reading the flow information of the multi-agent system stored in the storage server according to the information of the interaction request to obtain the second flow information, may specifically include the step: S241, obtaining the first action intention information in the information of the interaction request from the storage server, and reading the flow information of the multi-agent system stored in the storage server according to the first action intention information to obtain the second flow information including relevant information of the first object; for example, for the task of the agent eating an apple, the second flow information may include information about the scene where the apple is located.

[0104] The aforementioned step S250, i.e., utilizing the management agent computing service system to calculate the operating tasks of the environment in the multi-agent system based on the second flow state information and the acquired information about the interaction request to obtain an environment operating result, may specifically include the following step: S2521, utilizing the management agent computing service system to calculate the operating tasks of the environment in the multi-agent system based on the first action intention information and information related to the first object to obtain an environment operating result including a change result of the first object in the environment. For example, for a task in which an agent eats an apple, the environment operating result may be the result of deleting the eaten apple from the environment.

[0105] The above-mentioned step S260, i.e., storing the environment operation result in the storage server according to the distributed consistency storage protocol, may specifically include the step: S261, storing the information corresponding to the change result in the storage server according to the distributed consistency storage protocol.

[0106] Step S270, i.e., the first agent reading the information corresponding to the environment operation result from the storage server, may specifically include the following step: S271, the first agent reading information corresponding to the change result of the first object in the environment from the storage server; the information corresponding to the environment operation result includes information corresponding to the change result of the first object in the environment. For example, in a task in which the agent eats an apple, the information corresponding to the change result of the first object in the environment may include the uneaten apple.

[0107] The above-mentioned step S280, i.e., the first agent obtains the state change information of the first agent using the computing server and the information corresponding to the operation result of the environment, may specifically include the step: S281, the first agent determines the state change information of the first agent using the computing server and the information based on the change result of the first object in the environment. For example, in the task of the agent eating an apple, the agent finds that the apple to be eaten is no longer in the environment and realizes that the apple has been eaten. In other embodiments, the state change information of the first agent can be obtained by the agent or the management agent based on the change result of the first object, such as the hunger and thirst state of the first agent.

[0108] The aforementioned step S290, i.e., storing the state change information of the first agent in the storage server according to the distributed consistency storage protocol, includes: S291, storing the change information of the first category of the state of the first agent in the storage server according to the distributed consistency storage protocol. For example, in a task where the agent eats an apple, the information indicating that the first agent has eaten the apple may be stored in the storage server.

[0109] In a more specific embodiment, the above-mentioned step S2521, that is, using the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system based on the first action intention information and the relevant information of the first object to obtain the environment operating results including the change results of the first object in the environment, may more specifically include the steps: S25211, locally obtaining the corresponding environment evolution law information from the management agent computing service system according to the first action intention information; S25212, using the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system based on the first action intention information, the relevant information of the first object, and the environment evolution law information to obtain the change results of the first object in the environment; S25213, using the management agent computing service system and calculating according to the change results of the first object in the environment to obtain the state change information of the first agent; the environment operation results include the change results of the first object and the change information of the first category state of the first agent.

[0110] In this case, the aforementioned step S260, i.e., storing the environmental operation results in the storage server according to the distributed consistency storage protocol, may further include the step of storing the change information of the first category state of the first agent in the storage server according to the distributed consistency storage protocol. The aforementioned step S270, i.e., the first agent reading the information corresponding to the environmental operation results from the storage server, may further include the step of the first agent reading the change information of the first category state of the first agent from the storage server.

[0111] Furthermore, the aforementioned step S281, i.e., the first agent determining the state change information of the first agent using the computing server and information about the change result of the first object in the environment, may specifically include the step: S2811, wherein the first agent determines the change information of the second category state of the first agent using the computing server and information about the change result of the first object in the environment and the change information of the first category state of the first agent. The aforementioned step S290, i.e., storing the state change information of the first agent to the storage server according to the distributed consistency storage protocol, may also include the step: S291, storing the change information of the second category state of the first agent to the storage server according to the distributed consistency storage protocol. The change information of the first category state of the first agent may be information about changes in the ability state or personality state. For example, for the task of the agent eating an apple, it may be information about changes in the first agent's hunger level. The change information of the second category state of the first agent may be information about changes in the cognitive state or knowledge state, such as the agent recognizing that it has eaten an apple.

[0112] In these embodiments, the management agent computing service system can not only calculate the evolution of the computing environment, but also calculate part of the state change information of the agent. The agent can use the computing server to calculate another part of the state change information.

[0113] In other specific embodiments, the operation method of the multi-agent system described in each embodiment, for example, Figure 2 The method shown may further include the following steps: S2100, the first agent in the multi-agent system reads the external state information of the second agent in the multi-agent system stored in the storage server; S2110, the first agent uses the computing server and calculates the mental modeling task in the running task of the first agent based on the external state information of the second agent to obtain an interaction request about the mental information; S2120, stores the interaction request about the mental information to the storage server according to the distributed consistency storage protocol. In this embodiment, the computing server and the storage server can be used to implement the mental modeling of one agent on other agents. The information required for mental modeling can be read from the storage server, and the calculation process of mental modeling can be implemented on the computing server. For other agents, taking the second agent as an example, the external state information of the second agent may include the behavior information of the second agent.

[0114] In some other specific embodiments, the operation method of the multi-agent system described in each embodiment, for example, Figure 2The method shown may also include the following steps: S2130, the second agent in the multi-agent system reads the flow information of the multi-agent system stored in the storage server to obtain third flow information; S2140, the second agent uses the computing server and calculates the running task of the second agent based on the third flow information to obtain the interaction request of the second agent; S2150, stores the interaction request information of the second agent to the storage server according to the distributed consistency storage protocol; S2160, obtains the interaction request information of the second agent from the storage server, and reads the flow information of the multi-agent system stored in the storage server according to the interaction request information of the second agent to obtain fourth flow information.

[0115] In this case, the above-mentioned step S250, i.e., using the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the second flow information and the obtained information of the interaction request to obtain the environment operation result, may specifically include the steps of: using the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the second flow information, the fourth flow information, the information of the interaction request of the first agent and the information of the interaction request of the second agent to obtain the environment operation result.

[0116] Furthermore, the operation method of the multi-agent system described in each embodiment may also include the steps: S2170, the second agent reads the information corresponding to the environment operation result from the storage server; S2180, the second agent uses the computing server and the information corresponding to the environment operation result to obtain the state change information of the second agent; S2190, stores the state change information of the second agent to the storage server according to the distributed consistency storage protocol.

[0117] In these embodiments, the relevant steps of the second agent can be similar to the relevant steps of the first agent. The main difference lies in the calculation content of the management agent computing service system, which is calculated based on the relevant information of multiple agents, such as the interaction request information and flow information of the first agent and the second agent. In this way, the computing tasks when multiple agents act together on the environment can be completed. For example: two agents chase each other in the game of "tearing name tags". Who can tear it off and who can dodge it can be determined and changed by the management agent (world agent) based on the real-time status information of the two agents. For another example, in games such as "grabbing the stool", the control of the stool can also be determined by the management agent.

[0118] In addition, an embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method for operating a multi-agent system described in any of the above embodiments are implemented.

[0119] The present invention will be described below with reference to specific embodiments to facilitate a better understanding of the present invention.

[0120] In order to improve the versatility of multi-agent systems, especially for the needs of multi-agent collaborative interaction, mental modeling and dynamic entry and exit, a distributed general artificial intelligence agent system architecture is proposed. Based on this architecture, the multi-agent environment will be easier to build and have better stability and elastic expansion capabilities. The proposed distributed structure mainly includes three main structures: (1) distributed storage system (world server); (2) general artificial intelligence individuals (agents); (3) general artificial intelligence computing services (management agents). The world server can perform distributed consistent storage, the agent individuals (i.e., agents) can be provided with computing power by the computing servers in the distributed servers, and the management agents (i.e., system agents) can be provided with computing power by the containerized computing modules. For the logical relationship between the various components, see Figure 3 The agent can send action requests or event issuances to the world server. The world server can send the agent's individual request or event calculation task to the management agent. The management agent can return the environment evolution results or agent interaction results to the world server after calculation. The world server can return the scene perception results / action results to the agent.

[0121] The data structure of an intelligent entity may include: world identifier, entity identifier, entity artificial intelligence model, entity capabilities, value, knowledge, etc. For example, the data structure of an intelligent entity is:

[0122] WordID:0x0001

[0123] ID:Sally

[0124] Model: AOG

[0125] Ability: {pick, drop, observe}

[0126] Value: {safe, curious, clean}

[0127] Knowledge: {common sense, basic physics}

[0128] The data structure of the above-mentioned intelligent agent can be set on the computing server.

[0129] The data structure of the world server (storage server) may include: world ID, list of item IDs in the scene, list of agent IDs, list of current item states, list of current agent states, agent individual action / event requests, and agent individual action / event results. For example, the data structure of the world server is as follows:

[0130] World ID: World 0x0001

[0131] Item ID List

[0132] List of current status of items

[0133] Foo:{xxx}

[0134] Bar:{xxx}

[0135] List of current status of individuals

[0136] Sally:{xxx}

[0137] Anne:{xxx}

[0138] Individual Action / Event Application

[0139] Anne: {pick Foo}

[0140] Sally: {look for Bar}

[0141] Individual action / event results

[0142] [Foo picked up]

[0143] [Bar could not be seen by Sally]

[0144] The data structure of the agent and the data structure of the storage server store the corresponding world ID and agent ID (such as Sally), so that the data information is consistent.

[0145] The data structure of the management agent may include: world type, world rules, and interactive computing task list. For example, the data structure of the management agent may be as follows:

[0146] World Type: Simulated Indoor

[0147] World Laws: {Liquid Flow, Gravity, Object Breakage, Visibility Occlusion}

[0148] Interactive computing task list: {is it feasible for [Anne to pick Foo], [can Sally see Bar]}

[0149] In this embodiment, the general artificial intelligence agent can be driven by a variety of models, such as reinforcement models, large language models, AND / OR graph models, Markov models, etc. At the same time, the relevant models can contain other data structures unique to the agent model. For example, the AOG model uses content such as the capability part, the value part, and the knowledge part. The capability part represents the actions or interactions that the agent can issue to the world server, and also includes the content perception and joint understanding process of the world scene. The value part is the fundamental basis for the agent to make decisions, and the execution path decision is made based on the possible results of the capability and the value setting. The knowledge part is the inherent common sense of the world and the experience gained through the learning process, which is used to better optimize the path and understand the results during the decision-making process.

[0150] In the world server, various information about the corresponding world can be stored and recorded in a key-value format, which can be called a flow or flow information. The flow primarily includes information about agents and objects in the scene. This information is represented using an ontology and remains consistent across all individual agents and management agents. For example, concepts such as "Company," "Employee," and "Project" in the ontology are considered "concepts," while specific companies such as "Widgets.Inc." and "Bob," an employee, are considered instances. Instances can have contractual relationships, superior-subordinate relationships, reporting relationships, and collaborative work relationships.

[0151] These descriptions can be written in standard ontology languages ​​(such as OWL2), which can be stored in a key-value format. The most important aspect of managing intelligent agents is computational services. This separation from the traditional "world" and "environment" is, on the one hand, because we want the state-carrying part to be independent. This way, the implementation of the state part will be simple and efficient enough to support high concurrency requirements. On the other hand, the state evolution and computation itself is highly dependent on the number and complexity of intelligent agents in the world. By adopting a resource pool approach, we can better achieve elastic expansion of computing power.

[0152] It is worth noting that the decision-making and calculation of general intelligent agents also require a lot of computing power. Figure 4The intelligent agent (computing server) and the world server can communicate data related to events and actions. The management agent (management agent computing service module) and the world server can communicate data related to evolution and laws. Multiple world servers (storage servers) provide distributed and consistent storage for the information required by the intelligent and management agents. It can be seen that intelligent agents and management agents share many similarities in nature: both require significant computational resources, the amount of computation is highly variable, and both require access to streaming data from the world server. The main difference between the two is that general intelligent agents primarily read and understand the state of streaming data and write a small number of change requests to that data. In this case, the action is, for example, the action the agent is performing. The "small number" here refers primarily to change requests related to the agent itself, without considering the environmental state updates caused by the consequences of a large number of actions. Management agents primarily read change requests for streaming data and write a large number of state results related to that data.

[0153] As you can see, the distributed world servers store world information using a distributed consistency protocol (such as Paxos or Raft). This design is driven by the fact that the world service itself requires high stability and multi-client access. This approach ensures highly fault-tolerant content consistency in complex asynchronous scenarios (such as slowness, unresponsiveness, disconnections, and partial availability). Furthermore, its distributed nature and dynamic join and exit capabilities ensure both high availability and low maintenance costs.

[0154] The connection between individual agents and world servers is flexible. Based on physical proximity and communication bandwidth, they can freely choose to connect to different physical servers. Interaction is based on events and actions, specifically implemented through a Pub / Sub (subscribe / publish) communication queue. Information about individual agents in the world is synchronized across different regions through distributed storage, enabling real-time roaming and transaction continuity.

[0155] In traditional world evolution, the management agent (the traditional "environment" or "world") plays a nearly equal role in our processing process as individual agents. This equivalence is designed to ensure dynamic additivity and automatic load balancing of computing power. The management agent also uses message queues to manage the evolution of multiple worlds on the world server, processing them in real time through a computing resource pool. Separating the computation of world evolution laws from the data stored in the world is key to ensuring elastic computing power growth.

[0156] This newly designed distributed architecture offers several beneficial technical benefits. First, it isolates the environment / world data structure. By designing an ontology-based data structure, world data is stored in a distributed storage system based on distributed consensus (e.g., etcd / Zookeeper). Second, it eliminates the specificity of the system environment. In this distributed architecture, the system and individual agents are architecturally equivalent, differing primarily in the rights and visibility to modify objects in the environment. Environmental changes perceived by individual agents are essentially modifications to world data by the "system agent" (management agent). Individual actions on the environment effectively become "modification intentions" to the world data. Third, it enhances overall asynchrony. Since changes and interactions with the world are based on modifications to the distributed storage world structure, the atomicity and consistency of this process rely on the characteristics of the distributed consensus system, eliminating the need for "world synchronization" to update data. Fourth, it significantly enhances scalability. The distributed nature of the world processing solves the computational bottleneck that previously existed. This allows for parallel computing power by dynamically adding data to the "system agents" and dividing and conquering their processing content. This enables elastic scalability without downtime, supporting arbitrarily complex world logic and any number of agents.

[0157] In this embodiment, the world data design is based on distributed storage. The world environment is equivalent to the agent, and the real world and the ontological world data are updated synchronously. Decentralized distributed storage is used as the core of the world data, which can carry out dynamic asynchronous environment interaction and interaction between agents. Agents can be dynamically added and deleted, and individual capabilities can be dynamically changed. The computing power elastic expansion and contraction mechanism supports worlds and agent individuals of arbitrary complexity, and has strong correspondence with the real world. The multi-agent system architecture design based on distributed consistent storage can significantly improve the computing bottleneck problem of the system when the number of multi-agents is large. At the same time, it can realize communication interaction in an asynchronous manner, realize dynamic adjustment of the number of agents in the system and the ability characteristics of the agents, and realize good system scalability. It can solve the problems of performance limitations and poor expansion capabilities of world environment construction in traditional solutions.

[0158] In the description of this specification, the description with reference to the terms "one embodiment", "a specific embodiment", "some embodiments", "for example", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in an appropriate manner in any one or more embodiments or examples. The order of steps involved in each embodiment is used to schematically illustrate the implementation of the present invention, and the order of steps therein is not limited and can be appropriately adjusted as needed.

[0159] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0160] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0161] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0162] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0163] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-agent system operating device, characterized in that: include: A distributed storage system comprising at least one storage server; All of the storage servers are used to store flow information of the multi-agent system according to a distributed consistency storage protocol; the flow information includes agent information and scene information in the environment; An agent computing system comprising at least one computing server; at least one computing server being configured to compute an operating task of at least one agent in the multi-agent system based on flow information in the storage server; A management agent computing service system is used to calculate the operating tasks of the environment of the multi-agent system based on the flow information in the storage server to obtain the environment operating results; all the storage servers are also used to store the environment operating results according to the distributed consistency storage protocol.

2. The operating device of the multi-agent system according to claim 1, characterized in that: Also includes management server; The storage server is further configured to receive an instruction for an agent to enter the multi-agent system, and the management server is further configured to allocate a computing server to perform calculations on the running tasks of the entering agent; and / or The storage server is further configured to receive an instruction for an agent to exit the multi-agent system, and the management server is further configured to stop the corresponding computing server from computing the running tasks of the exiting agent; and / or The management server is also used to adjust the computing server in the intelligent computing system to the storage server in the distributed storage system, or to adjust the storage server in the distributed storage system to the computing server in the intelligent computing system according to the computing power of the intelligent computing system and the storage situation of the distributed storage system.

3. The operating device of the multi-agent system according to claim 1, characterized in that: The management agent computing service system is also used to store the environmental evolution law information of the multi-agent system, and to calculate the operating tasks of the environment of the multi-agent system based on the flow information in the storage server and the environmental evolution law information.

4. The operating device of the multi-agent system according to claim 1 or 3, characterized in that: The management agent computing service system includes at least one management agent computing service module, and the management agent computing service system is a distributed computing system; and / or, The intelligent agent computing system is a distributed computing system.

5. The operating device of the multi-agent system according to claim 1 or 3, characterized in that: The management agent computing service system is a distributed computing system, and the management agent computing service system includes at least one management agent computing service module, each of which is a containerized computing service module; and / or, Each of the storage servers and each of the computing servers is a physical server.

6. The operating device of the multi-agent system according to claim 1, characterized in that: The running task of the agent includes at least one of a task of calculating the agent's action intention to the environment and a task of calculating the agent's mental modeling of other agents; and / or, The running tasks of the environment include at least one of the following: computing tasks for changes in the state of the environment caused by the actions of the intelligent agent, computing tasks for natural changes in the environment according to information on the laws of environmental evolution, and computing tasks for determining the results of the joint actions of multiple intelligent agents; and / or, The agent information includes the agent's state information, the agent's action intention information, and the agent's action result information; the scene information in the environment includes the state information of objects in the scene in the environment; and / or, Each of the intelligent agents includes a predetermined artificial intelligence agent model and an intelligent agent data structure.

7. The operating device of the multi-agent system according to claim 1 or 2, characterized in that: The agents in the multi-agent system communicate asynchronously with the corresponding computing servers through subscription / publishing; and / or, The management agent computing service system and the corresponding operation server realize asynchronous communication through subscription / publishing.

8. A method for operating a multi-agent system, characterized in that: The operating device of the multi-agent system according to any one of claims 1 to 7, wherein the operating method comprises: The first agent in the multi-agent system reads the flow state information of the multi-agent system stored in the storage server to obtain first flow state information; The first agent calculates the running task of the first agent using the computing server and according to the first flow state information to obtain an interaction request; Storing the information of the interaction request in a storage server according to a distributed consistency storage protocol; Acquire the interaction request information from the storage server, and read the flow state information of the multi-agent system stored in the storage server according to the interaction request information to obtain second flow state information; Utilizing a management agent computing service system and calculating an operating task of an environment in the multi-agent system according to the second flow state information and the acquired information of the interaction request, to obtain an environment operating result; Storing the environment operation result in a storage server according to the distributed consistency storage protocol; The first agent reads information corresponding to the environment operation result from the storage server; The first agent obtains the state change information of the first agent by using the computing server and information corresponding to the environment operation result; The state change information of the first agent is stored in a storage server according to a distributed consistency storage protocol.

9. The method for operating a multi-agent system according to claim 8, wherein: Utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the second flow state information and the acquired information of the interaction request, to obtain an environment operating result, including: The management agent computing service system is used to calculate the operating tasks of the environment in the multi-agent system based on the second flow information, the information of the interaction request obtained, and the environment evolution law information stored locally in the management agent computing service system to obtain the environment operation results.

10. The method for operating a multi-agent system according to claim 9, wherein: The management agent computing service system is used to calculate the environment operation task in the multi-agent system based on the second flow state information, the obtained information of the interaction request, and the environment evolution law information stored locally in the management agent computing service system to obtain the environment operation result, including: Utilizing multiple management agent computing service modules in the management agent computing service system and based on the second flow information, the information obtained about the interaction request and the environment evolution law information stored locally in the management agent computing service system, distributed computing is performed on the environment operation tasks in the multi-agent system to obtain the environment operation results.

11. The method for operating a multi-agent system according to claim 10, wherein: The first agent calculates the running task of the first agent using the computing server according to the first flow state information to obtain an interaction request, including: The first agent uses multiple computing servers in the agent computing system and performs distributed computing on the running task of the first agent according to the first flow state information to obtain an interaction request; The first agent obtains the state change information of the first agent by using the computing server and information corresponding to the environment operation result, including: The first agent utilizes multiple computing servers in the agent computing system and information corresponding to the environment operation results to perform distributed computing on the operation tasks of the first agent to obtain state change information of the first agent.

12. The method for operating a multi-agent system according to claim 8, wherein: Also includes: increasing or decreasing the number of computing servers in the agent computing system according to the computing power of the agent computing system; and / or, Increase or decrease the number of storage servers in the distributed storage system according to the storage situation of the distributed storage system; and / or, According to the computing power of the intelligent computing system and the storage situation of the distributed storage system, some of the computing servers in the intelligent computing system are adjusted to storage servers of the distributed storage system, or, some of the storage servers in the distributed storage system are adjusted to computing servers of the intelligent computing system.

13. The method for operating a multi-agent system according to claim 8, wherein: The first agent in the multi-agent system reads the flow state information of the multi-agent system stored in the storage server to obtain first flow state information, including: the first agent in the multi-agent system queries the flow state information of the multi-agent system stored in the storage server to obtain current information of a first object in the environment of the multi-agent system; the first flow state information includes the current information of the first object; The first agent calculates the first agent's operation task using the computing server and based on the first flow information to obtain an interaction request, including: the first agent calculates the first agent's operation task using the computing server and based on the current information of the first object to obtain first action intention information of the first agent with respect to the first object; the interaction request includes the first action intention information; Storing the information of the interaction request in a storage server according to a distributed consistency storage protocol includes: storing the first action intention information in a storage server according to a distributed consistency storage protocol; Obtaining the interaction request information from the storage server, and reading the flow information of the multi-agent system stored in the storage server according to the interaction request information to obtain second flow information, including: obtaining the first action intention information in the interaction request information from the storage server, and reading the flow information of the multi-agent system stored in the storage server according to the first action intention information to obtain second flow information including relevant information of the first object; Utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the second flow state information and the obtained information of the interaction request to obtain an environment operating result, including: utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the first action intention information and relevant information of the first object to obtain an environment operating result including a change result of the first object in the environment; Storing the environment operation result in a storage server according to the distributed consistency storage protocol, including: storing information corresponding to the change result in the storage server according to the distributed consistency storage protocol; The first agent reads information corresponding to the environment operation result from the storage server, including: the first agent reads information of a change result of the first object in the environment from the storage server; the information corresponding to the environment operation result includes information of a change result of the first object in the environment; The first agent obtains the state change information of the first agent using the computing server and information corresponding to the result of the operation of the environment, including: the first agent determines the state change information of the first agent using the computing server and information corresponding to the result of the change of the first object in the environment; Storing the state change information of the first agent in a storage server according to a distributed consistency storage protocol includes: storing the change information of the first category state of the first agent in a storage server according to a distributed consistency storage protocol.

14. The method for operating a multi-agent system according to claim 13, wherein: Utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the first action intention information and the relevant information of the first object, obtaining an environment operating result including a change result of the first object in the environment, including: obtaining corresponding environment evolution law information locally from the management agent computing service system according to the first action intention information; utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the first action intention information, the relevant information of the first object, and the environment evolution law information, obtaining a change result of the first object in the environment; utilizing the management agent computing service system and calculating according to the change result of the first object in the environment, obtaining state change information of the first agent; the environment operating result includes the change result of the first object and the change information of the first category state of the first agent; storing the environment operation result in a storage server according to the distributed consistency storage protocol, further comprising: storing change information of the first category state of the first agent in the storage server according to the distributed consistency storage protocol; The first agent reads information corresponding to the environment operation result from the storage server, further comprising: the first agent reads change information of the first category state of the first agent from the storage server; The first agent determines the state change information of the first agent using the computing server and information about the change result of the first object in the environment, including: the first agent determines the change information of the second category state of the first agent using the computing server and information about the change result of the first object in the environment and the change information of the first category state of the first agent; Storing the state change information of the first agent in the storage server according to the distributed consistency storage protocol also includes: storing the change information of the second category state of the first agent in the storage server according to the distributed consistency storage protocol.

15. The method for operating a multi-agent system according to claim 8, wherein: The operation method further comprises: The first agent in the multi-agent system reads the external state information of the second agent in the multi-agent system stored in the storage server; The first agent calculates the mental modeling task in the running task of the first agent using the computing server and according to the external state information of the second agent, and obtains an interaction request regarding mental information; The interactive request regarding the mental information is stored in a storage server according to a distributed consistency storage protocol.

16. The method for operating a multi-agent system according to claim 8, wherein: The operation method further includes: the second agent in the multi-agent system reads the flow information of the multi-agent system stored in the storage server to obtain third flow information; the second agent uses the computing server and calculates the operation task of the second agent based on the third flow information to obtain the interaction request of the second agent; stores the interaction request information of the second agent in the storage server according to the distributed consistency storage protocol; obtains the interaction request information of the second agent from the storage server, and reads the flow information of the multi-agent system stored in the storage server according to the interaction request information of the second agent to obtain fourth flow information; Utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the second flow state information and the obtained information of the interaction request to obtain an environment operating result, including: utilizing the management agent computing service system and calculating the operating tasks of the environment in the multi-agent system according to the second flow state information, the fourth flow state information, the information of the interaction request of the first agent, and the information of the interaction request of the second agent to obtain an environment operating result; The operation method also includes: the second intelligent agent reads the information corresponding to the environment operation result from the storage server; the second intelligent agent uses the computing server and the information corresponding to the environment operation result to obtain the state change information of the second intelligent agent; and stores the state change information of the second intelligent agent to the storage server according to the distributed consistency storage protocol.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 8 to 16 are implemented.

Images