An agent-based model-predictive controller for chilled water plants using wireless sensor and actuator networks

Reduced manning requirements and other cost reducing measures have spurred interest in automation of engineering plants onboard naval combat vessels. Furthermore, automation may increase resiliency of shipboard engineering plants when compared to the current generation of manually configured plants. This paper presents a control algorithm and deployment strategy which supervises a ship's chilled water plant to control the temperature of thermal loads (e.g., air chillers, electrical components). Redundant computation and communication capabilities motivates the use of an agent-based controller (ABC) enabled through a peer-to-peer wireless network. In the architecture presented, each agent sits on one or more digraphs corresponding to the utility generated by the fluid exiting the chilled water plant at each discharge point. Each digraph is a component of the decentralized model-predictive controller (MPC). Performance of the proposed control architecture is tested in simulation, and is shown to approach the performance of an effective, but computationally exhaustive, centralized MPC.