Multiobjective Optimization of Building Heating System
To be feasible and acceptable from economical and societal point of view, future smart building applications must be cheap and simple from both the customer’s and service provider’s perspective. Although there are many state-of-the art approaches to smart heating control, they often assume an expensive installation of the controller inside the building. Therefore, we focus on a remote control approach, where the building is equipped with a minimal set of sensors, actuators and an interface which communicates with a remote control agent: these solutions can highly overload the communication channel in the classical control settings. We propose a long-horizon approach, where the remote control agent sends the control actions only once per couple of hours (e.g. 6/12/24). This obviously brings many problems that can make this approach unfeasible.
Concretely, we used a Simulink model that simulates the thermal behaviour of a real office building. This simulator considers the building divided into zones consisting of rooms with similar thermal and physical characteristics. The remote control agent is based on a data-driven black box model, that is a surrogate of the simulator. NSGA-II algorithm use such model to optimise temperature set-point of the zones so as to minimise thermal consumptions and maximise users comfort. Optimised temperature set-points are then applied to the simulator in order to evaluate the resulting energy savings and comfort.