Smart decentralized MPC for temperature control in multi–zone buildings

Abstract

This paper proposes a smart decentralized model predictive control (MPC) approach that uses artificial intelligence to predict the temperatures of neighboring rooms in a building and does not require communication among MPC controllers. A single floor four–zone building model is utilized to investigate if artificial intelligence–based predictors are able to yield accurate predictions that result in comparable performance with a distributed MPC algorithm, which requires communication exchange among controllers. The artificial intelligence techniques utilized in this paper are recurrent neural networks and the K–nearest neighbors algorithm. The results show that the developed smart decentralized MPC can perform as well as a centralized or distributed MPC without requiring a communication network between local MPC controllers. Also, it outperforms decentralized MPC with integral action in terms of the used heat input.