Design of Equipment Control System Based on Large Language Models with Intelligent Validator Agent

This study presents a comprehensive framework leveraging large language models (LLMs) for robust and accurate equipment control. The system integrates Retrieval-Augmented Generation (RAG) to provide LLMs with real-time contextual information retrieved from external knowledge sources. Furthermore, a Mixture-of-Agents (MoA) architecture is employed to harness the collective intelligence of multiple LLMs, enhancing the quality and reliability of generated control suggestions. The proposed intelligent Validator Agent, equipped with natural language processing capabilities and a feedback-correction mechanism, serves as the core component. The Validator Agent translates the LLM’s textual responses into actionable commands, validates them against real-time equipment status and predefined rules from a database, and corrects potential errors before execution. Performance evaluation demonstrates that although RAG and MoA improve accuracy, the Validator Agent’s correction loop remains essential. This research highlights the potential of combining LLMs with RAG, MoA, and intelligent agents to create highly accurate and reliable natural language-based equipment control systems.