Model Predictive Control for Buildings

Energy savings in buildings have gained a lot of attention in recent years. Most of the research is focused on the building construction or alternative energy sources in order to minimize primary energy consumption of buildings. By contrast, this thesis deals with an advanced process control technique called model predictive control (MPC) that can take advantage of the knowledge of a building model and estimations of future disturbances to operate the building in a more energy efficient way. MPC for buildings has recently been studied intensively. It has been shown that energy savings potential of this technique reaches almost 40 % compared to conventional control strategies depending on the particular building type. Most of the research results are, however, based on simulation studies subject to number of assumptions. On the contrary, the objectives of this thesis are i) evaluate MPC energy savings potential on a real building, ii) develop and evaluate an alternative MPC formulation for buildings that is less sensitive to model mismatch and weather forecast errors, iii) develop and evaluate an alternative MPC formulation that takes into account mathematical formulas for thermal sensation of occupants. First of all, this thesis deals with the implementation of the MPC controller on a pilot building of Czech Technical University (CTU) in Prague. The development of a grey-box thermodynamical model for control, the formulation of the underlying optimization problem and the development of the software platform for optimization problem solving and communication of the optimal control moves to the building automation system are topics treated in detail. Moreover, the evaluation of the energy savings potential is provided, showing that for the investigated building, the savings are between 15 % and 28 %, power peak demand was lowered by 50 %, while the thermal comfort in the building was kept on a higher level. Then this thesis presents a tool that was used for the development of the MPC controller applied for the CTU building. The tool enables tuning and debugging of MPC controllers for buildings and allows users to explore controller behavior for different scenarios (e.g. weather conditions, occupancy profiles or comfort regimes). Afterwards, based on the assessment of the long term operation of the MPC controller applied to the control of the building of the CTU, the main issues for practical applicability of MPC are pointed out and an alternative optimal control problem formulation tackling the issues is proposed showing a better closed-loop performance even in situations when there is a model mismatch or disturbance prediction errors when comparing the performance to the formulations presented in the literature. Finally, this thesis deals with the development of a computationally tractable method for solving an alternative MPC problem formulation, which incorporates thermal comfort index predicted mean vote and which leads to a general constrained optimization problem. The advantage of this formulation is that it implicitly contains user perception of the thermal comfort in the cost function and thus it is possible to achieve better thermal comfort even with less input energy.