プロセス制御におけるマルチエージェントシステムの合意形成手法

抄録

In this paper, we suppose an agent network system in a multi-agent environment for process control and propose a calculation method of agent credibility, which is computed by aggregating other agents' evaluations. Two models called “center-of-gravity model” and “reflection model” are presented as the calculation methods. A test simulation was performed using a simple tank system, where the tank level is to be kept at a designed level. When an agent sends wrong data to other agents, the recipients calculate the credibility of the sender to identify faulty agents. Diagnosis agents identify a faulty condition by evaluating credibility of other agents based on constraints related to the task process. A tank agent has been introduced to distinguish the tank failure from agent failures. The results showed agents identified faulty agents while the controller agent maintained the tank level without any problems. Agents also detected the tank failure through decline of the credibility of the tank agent.