基于大语言模型与Petri网的智能体服务组合方法研究

Agent Service Composition Method Based on Large Language Models and Petri Nets

摘要: 随着大语言模型、智能体与服务计算技术的快速发展，如何高效、可靠地组合多个智能体服务成为研究热点。Petri网因其严格的数学表达与强大的行为分析能力，被广泛用于服务组合建模与验证。然而，传统Petri网合成方法依赖人工定义语义匹配规则，缺乏自动化与智能化支持。本文提出一种结合大语言模型与Petri网的智能体服务组合方法。首先，提出面向智能体服务编排的UML建模方法，统一描述服务流程；其次，设计UML模型向Petri网转换机制，利用XMI与XSLT实现自动化映射；进一步，引入大语言模型对Petri网进行语义标注与智能融合，实现基于语义的Petri网共享合成；最后，通过PIPE (Platform Independent Petri Net Editor)工具对合成后的Petri网进行模型检测，验证其行为正确性。实验结果表明，该方法能够有效提升智能体服务组合的自动化程度与可靠性。

Abstract: With the rapid development of Large Language Model (LLM), Agent and service computing technologies, how to efficiently and reliably compose multiple agent services has become a research hotspot. Petri nets are widely used for service composition modeling and verification due to their strict mathematical expression and strong behavioral analysis capabilities. However, traditional Petri net synthesis methods rely on manually defined semantic matching rules, lacking automation and intelligent support. This paper proposes an agent service composition method that combines large language models with Petri nets. First, a UML modeling method for agent service orchestration is proposed to uniformly describe service processes. Second, an XML-based mechanism for transforming UML models into Petri nets is designed, using XMI and XSLT to achieve automated mapping. Furthermore, large language models are introduced to semantically annotate and intelligently fuse Petri nets, enabling semantic-based shared synthesis of Petri nets. Finally, the synthesized Petri net is validated for behavioral correctness through model checking using the PIPE tool. Experimental results show that this method can effectively enhance the automation and reliability of agent service composition.