CN-122027542-A - Cross-domain agent discovery routing method and system for agent network

CN122027542ACN 122027542 ACN122027542 ACN 122027542ACN-122027542-A

Abstract

The invention provides a method and a system for discovering and routing a cross-domain intelligent agent facing an intelligent agent network, which constructs a cross-domain routing system composed of a multi-stage discovery server and a mutually trusted list, when a discovery server of a domain of a personal assistant intelligent agent does not find a required service intelligent agent in a current domain, a chained forwarding or aggregation forwarding strategy is adopted by an upper layer discovery server, a discovery request is protocol-diffused to a downstream trusted domain under the constraint of a forwarding upper limit and a mutually trusted range, and path tracing and step-by-step feedback are realized by means of an identity identifier of an accessed server. The method breaks through the capability boundary of single domain discovery, realizes the collaborative search of agents crossing manufacturers and management domains, ensures the safety and controllability of the cross-domain diffusion process through a mutually trusted list and forwarding constraint, avoids the out-of-control request, and simultaneously provides a traceable basis for the follow-up audit, charging and policy optimization by a complete path recording mechanism.

Inventors

YU KE
LIU JUN
GUO XIAOLIAN
SONG HAOZHE
LI YINMING

Assignees

北京邮电大学

Dates

Publication Date: 20260512
Application Date: 20260305

Claims (10)

1. An agent network-oriented cross-domain agent discovery routing method is characterized in that the method is executed based on a lower layer discovery server in an agent network, the agent network is composed of a plurality of independent discovery domains, a plurality of service agents are deployed in each discovery domain, one or more lower layer discovery servers in a domain are maintained by an upper layer discovery server for accessing personal assistant agents, and the method comprises the following steps: acquiring an agent discovery request of the personal assistant agent, and extracting a capability description text, a filtering condition and constraint parameters for constraining a cross-domain diffusion range of the required agent from the agent discovery request; According to the capability description text and the filtering condition, searching in the service intelligent agent deployed in the current discovery domain, if a candidate service intelligent agent meeting the condition is found, sending a discovery result to the personal assistant intelligent agent, and ending the searching, if the candidate service intelligent agent meeting the condition is not found, starting a cross-domain discovery operation, forwarding the intelligent agent discovery request to the upper layer discovery server corresponding to the current discovery domain, selecting a downstream discovery domain and a downstream upper layer discovery server corresponding to the downstream discovery domain from a mutually trusted list maintained locally according to the constraint parameter by the upper layer discovery server according to a chain forwarding or aggregation forwarding mode, forwarding the intelligent agent discovery request and the accessed upper layer discovery server identity to the downstream upper layer discovery server, calling a lower layer discovery server in the downstream discovery domain, searching the required intelligent agent, if the candidate service intelligent agent meeting the condition is found, feeding back to the initial personal assistant, and continuing to perform the cross-domain forwarding operation until the candidate service intelligent agent meeting the condition is not found; the chained forwarding is to diffuse to the downstream discovery domain searching along a single path hop by hop, and the aggregation forwarding is to synchronously diffuse to a plurality of downstream discovery domain searching by selecting a plurality of paths at each hop.
2. The agent network-oriented cross-domain agent discovery routing method of claim 1, wherein the personal assistant agent extracts keywords related to capability requirements and constraint requirements for users based on natural language requirement specifications based on a pre-trained large language model to fill a set template to obtain the capability description text, the filtering condition and the constraint parameters, and searches a preset agent performance standard knowledge base in combination with the keywords to supplement missing fields in the capability description text, the filtering condition and the constraint parameters and to search revision error fields.
3. The agent network oriented cross-domain agent discovery routing method of claim 1 wherein the capability description text includes descriptions of the technical field, required functions, required interface criteria, and invoked external tools; The filtering conditions comprise quality of service and performance index requirements, cost requirements, security attribute requirements, vendor requirements and deployment environment requirements; the constraint parameters include a maximum concurrent branch number and a forwarding upper bound for cross-domain discovery operations.
4. The agent network oriented cross-domain agent discovery routing method of claim 3, wherein the method cross-domain discovery operation further comprises: The accessed upper layer discovery server identity is recorded successively according to the access history to obtain forwarding path information, the forwarding path information, a forwardable mutually trusted server range and a check code for trust verification are forwarded to a downstream upper layer discovery server corresponding to the downstream discovery domain, and updating is carried out in each cross-domain discovery operation; updating the constraint parameters in the current downstream discovery domain by checking forwarding path information, if the candidate service agent meeting the conditions is found in the current downstream discovery domain, feeding back the candidate service agent to the initial personal assistant agent step by step according to the forwarding path information, and if the candidate service agent meeting the conditions is not found in the current downstream discovery domain, continuing to execute the cross-domain discovery operation in the downstream according to the updated constraint parameters until the forwarding upper limit is reached.
5. The agent network oriented cross-domain agent discovery routing method of claim 4 wherein the chained forwarding includes the steps that when the downstream upper layer discovery server does not find the candidate service agent meeting the condition in the current downstream discovery domain, an object which is not accessed in the mutually trusted server range is selected as a downstream upper layer discovery server of the next cross-domain discovery operation, and the forwarding upper limit in the constraint parameter is reduced by 1 until the value of the forwarding upper limit is 1, and the cross-domain discovery operation is ended; the aggregation forwarding includes that when the downstream upper layer discovery server does not find the candidate service agent meeting the conditions in the current downstream discovery domain, a plurality of unviewed objects are selected in the range of the mutually trusted servers to serve as downstream upper layer discovery servers of next cross-domain discovery operation, forwarding upper limits are respectively configured for each downstream upper layer discovery server, and the following conditions are met: ; Wherein, the M is the total number of downstream upper layer discovery servers in the next cross-domain discovery operation determined by the current upper layer discovery server i, And (5) a forwarding upper limit value corresponding to the current upper layer discovery server i in the current cross-domain discovery operation is obtained.
6. The agent network oriented cross-domain agent discovery routing method of claim 1, further comprising: the lower layer discovery server extracts high-dimensional semantic vectors from capability description texts and performance parameters of all service agents in the discovery domain by utilizing a pre-trained semantic extraction model, and stores the high-dimensional semantic vectors into a vector database; and responding to the agent discovery request, extracting target bit semantic vectors from the capability description text and filtering conditions of the required agent by using the semantic extraction model, and searching vector similarity in the vector database to inquire the candidate service agent meeting the conditions.
7. The method for discovering and routing an agent network-oriented cross-domain agent according to claim 1, wherein the lower layer discovery server further establishes a cache layer for recording hot spot service agents with access frequencies higher than a set value and discovery paths thereof in a preset time window, when a new agent discovery request arrives, the cache layer is firstly queried preferentially, and if the hot spot agents meeting the conditions and the discovery paths corresponding to the hot spot agents exist in the cache layer, a result is directly returned.
8. The agent network oriented cross-domain agent discovery routing method of claim 1 further comprising recording, by the lower layer discovery server, an average response delay and a cross-domain discovery success rate of historically processing the agent discovery request, and selecting to perform a cross-domain discovery operation in a form of chained forwarding or aggregated forwarding according to a preset value range interval.
9. An agent network-oriented cross-domain agent discovery routing system, comprising a plurality of overall network discovery domains, each of which deploys a plurality of service agents, one or more lower-level discovery servers within a domain maintained by an upper-level discovery server for accessing personal assistant agents to perform agent discovery operations; the lower layer discovery server is configured to perform the agent network-oriented cross-domain agent discovery routing method according to any one of claims 1 to 8.
10. A computer program product comprising a computer program or instructions which, when executed by a processor, carries out the steps of the method according to any one of claims 1 to 8.

Description

Cross-domain agent discovery routing method and system for agent network Technical Field The present invention relates to the technical field of intelligent agent networks, and in particular, to a method and a system for discovering and routing a cross-domain intelligent agent facing an intelligent agent network. Background With the rapid development of artificial intelligence technology, agents (agents) have evolved from single individuals performing fixed tasks toward multi-Agent collaboration and cross-platform interconnection. The concept of the internet (Internet of Agents, ioA) of agents has been developed, aiming at connecting agents scattered in different manufacturers and different management domains through standardized communication protocols to form a dynamic collaboration network capable of self-organizing and self-negotiating so as to break through the capability limitation of a single agent and meet the increasingly complex task demands of users. In the intelligent agent internet architecture, how to efficiently and accurately discover service intelligent agents with specific skills is the basis for realizing task allocation and collaborative execution. Currently, the main current agent discovery schemes in the industry focus mainly on centralized management within a single domain. In particular, the prior art generally employs 1) a service registry or agent directory is deployed within a single platform or organization based on the service registry's domain discovery mechanism, with the service agent registering its capability description, interface address, and state information with the center at start-up. When the personal assistant agent needs to cooperate, a query request is initiated to the center, and the center matches and returns an agent list meeting the conditions in the local domain. Such mechanisms are common in closed environments or single vendor ecologies, such as part of the cloud vendor provided smart markets or tool call platforms. 2) And establishing an index database of the intelligent capability in the domain based on the capability index retrieval method, and supporting retrieval based on keywords, skill labels or interface types. When a task request is received, the system searches the service agents with matched capabilities in the local domain index library by analyzing the task requirement, and selects a proper agent to execute the task through a preset scheduling strategy. However, the prior art has the disadvantage in practical application that, firstly, the core limitation of the prior art is to find the closure of the scope. The management scope of the service registry or the capability index library is limited to a single domain, so that information islands are formed among agents of different manufacturers and different management domains. When no service agent meeting the requirements exists in the local domain, the prior art lacks an effective cross-domain query mechanism, can not utilize the agent resources of other domains, and limits the service capability of the whole system. Second, there is a lack of a agreed cross-domain routing mechanism. Even if there is a need for cross-domain invocation, existing schemes often rely on manually configured fixed routing or static whitelist interfacing, lacking standardized, dynamically extensible discovery protocols. This approach is complex to configure, difficult to maintain, and does not allow flexible adjustment of the discovery path according to network status or task requirements. Moreover, trust and security issues in cross-domain scenarios are not fully considered. In a multi-vendor, multi-administrative domain environment, there may be no default trust relationship between the different domains. The prior art does not provide an access control mechanism based on a mutual trust relationship, and is difficult to ensure that discovery requests are only spread in a trusted range, so that potential safety hazards exist. Finally, the discovery process lacks controllability and traceability. The existing scheme mostly adopts a simple broadcasting or traversing mode when cross-domain inquiry is carried out, lacks effective constraint on forwarding depth and the number of concurrent branches, and is easy to cause network resource waste or out-of-control request. Meanwhile, due to the lack of a complete path record, when the discovery result is abnormal or needs to be examined, the discovery process cannot be effectively traced. Therefore, the current agent discovery technology is mainly oriented to a single-domain scene, and a set of general discovery schemes suitable for cross-manufacturer, cross-management-domain, protocol routing support and safe and controllable diffusion are not formed. The technical blank limits the large-scale application and cross-ecological cooperation capability of the intelligent agent Internet, and a new intelligent agent discovery technical scheme is needed. Disclosure of Invention In