A multi-agent MPC architecture for distributed large scale systems

Abstract

In the present work, techniques of Model Predictive Control (MPC), Multi Agent Systems (MAS) and Reinforcement Learning (RL) are combined to develop a distributed control architecture for Large Scale Systems (LSS). This architecture is multi-agent based. The system to be controlled is divided in several partitions and there is an MPC Agent in charge of each partition. MPC Agents interact over a platform that allows them to be located physically apart. One of the main new concepts of this architecture is the Negotiator Agent. Negotiator Agents interact with MPC Agents which share control variables. These shared variables represent physical connections between partitions that should be preserved in order to respect the system structure. The case of study, in which the proposed architecture is being applied and tested, is a small drinking water network. The application to a real network (the Barcelona case) is currently under development.