Conference Year



MPC, automation, multi-zone buildings, IAQ


Heating Ventilation and Air Conditioning (HVAC) represents a large fraction of the world’s primary energy demand. Novel control strategies such as Model Predictive Control (MPC) aim to reduce energy use, while also improving occupant comfort. For MPC to be a viable alternative to classical Rule Based Control (RBC), it should be able to incorporate multiple emission and production systems, multiple zones, and aspects such as thermal comfort and indoor air quality (IAQ). Identifying grey-box or black-box MPC controller models for hybrid energy systems in multi-zone buildings has proven to be difficult. White-box models use physical knowledge of the system and take into account the desired dynamics. This approach however requires a substantial time investment since every building requires a custom model. The goal of this paper is to describe on-going work aiming at an automated methodology for setting up MPC controllers for buildings using white-box models. For this methodology, firstly, a detailed ‘emulator’ model of the building needs to be developed. Secondly this emulator is linearised to obtain a state space formulation of the building system and thirdly the emulator is used to generate MPC input data such as disturbances. Fourthly, a custom MPC tool uses this information to compute optimal control set points. These steps are elaborated below. The first step of the methodology is to the IDEAS library in Modelica to set up a detailed building envelope model. Modelica is an object-oriented equation based language that allows assembling complex systems by combining component models from open source libraries such as IDEAS. The second step is to linearise the building model. The IDEAS library is parametrized such that non-linearities such as convection correlations and radiation can be linearised around a well-chosen working point. Hydraulic connections from the HVAC are simplified and converted into heat flow rates. The HVAC is therefore simplified such that their the heat flow rate set points can be optimized. Disturbances such as the ambient temperature are also model inputs. The state space model resulting from the linearisation, although linear, accurately predicts the temperatures of the building’s zones. In the third step the emulator model is used to compute and store time series data for all state space model inputs such as ambient and radiative temperatures, solar incidence on glazing and internal gains from occupants. In step four the highly accurate state space model (step two) and corresponding input data (step three) files are used to set up the MPC problem. This MPC optimizes the remaining inputs from the state space model, subject to constraints and using a cost function that are passed to the optimization problem using augmented rows of the state space matrices. The state space matrices are pre-processed using CasADi such that a computationally efficient linear program is generated. This methodology is demonstrated on a medium size office building with 32 zones and hybrid emission and production systems. Results and performance are discussed. The strong points are the applicability to hybrid energy systems in multi-zone buildings, allowing the evaluation of thermal comfort and IAQ in different zones.Â