Model Predictive Control for Building Automation

Abstract

We propose a building HVAC system, integrating local energy production and storage, together with a model based controller. The heating system integrates several local heat production and storage devices and multiple fluid circuits at different temperatures to minimize entropy production through mixing. The controller uses a model of the system and predictive knowledge of demand and weather information to minimize electrical energy import, while maintaining thermal comfort by solving mixed integer optimization problems online. Time-varying and unknown system parameters are estimated and adapted online, using an unscented Kalman filter. The adaptation greatly reduces modeling effort and maintenance cost. The proposed setup is tested in a co-simulation, using a physical (modelica-) model of the building and energy system as well as realistic weather and demand data. Our system delivers nearly seven times more energy in the form of heat, than it needs to import (electrical) energy from external sources.