王航 黄小平 甘国操 冉瑛
（中国民航信息网络股份有限公司重庆分公司，重庆 401122）
摘要：针对企业在数字化转型过程中，采用单一智能体框架实现多业务场景智能化时面临的困境，提出一种基于大型语言模型的异构多智能体系统。该系统基于企业业务领域划分智能体，并构建一种异构多智能体系统分层架构，使智能体系统以分层的方式协同工作。其中，主控智能体负责业务领域识别、服务注册、服务分发和结果汇集等工作；各领域智能体以微服务架构独立运行，提升了异构多智能体系统的效率和灵活性。通过实验验证了该系统在众多且差异化较大的场景下，不仅能够完成特定任务，还能够自主选择执行效果最好的智能体。
关键词：异构多智能体系统；大型语言模型；主控智能体；领域智能体；微服务架构
中图分类号：TP18 文献标志码：A 文章编号：1674-2605(2025)04-0003-06
DOI：10.12475/aie.20250403 开放获取
Heterogeneous Multi-agent System Based on Large Language Models
WANG Hang HUANG Xiaoping GAN Guocao RAN Ying
(China Travelsky Information Technology Co., Ltd., Chongqing Branch, Chongqing 401122, China)
Abstract: Aiming at the challenges faced by enterprises during digital transformation when adopting a single-agent framework to achieve intelligent solutions across multiple business scenarios, this paper proposes a heterogeneous multi-agent system based on large language models. The system categorizes agents based on enterprise business domains and constructs a hierarchical architecture for heterogeneous multi-agent systems, enabling agents system to collaborate in a layered manner. Within this architecture, a master agent is responsible for tasks such as business domain recognition, service registration, service distribution, and result aggregation. Domain-specific agents operate independently based on a microservices architecture, enhancing the efficiency and flexibility of the heterogeneous multi-agent system. Experimental results demonstrate that the system not only accomplishes specific tasks across numerous and highly diverse scenarios but also autonomously selects the agent with the best execution performance.
Keywords: heterogeneous multi-agent system; large language models; master agent; domain-specific agents; microservices architecture
