Optimal control, State estimation, White-box model, Energy performance of buildings
In order to improve the energy efficiency of buildings, optimal control strategies, such as model predictive control (MPC), have proven to be potential techniques for intelligent operation of energy systems in buildings. However, in order to perform well, MPC needs an accurate controller model of the building to make correct predictions of the building thermal needs (feedforward) and the algorithm should ideally use measurement data to update the model to the actual state of the building (feedback). In this paper, a white-box approach is used to develop the controller model for an office building, leading to a model with more than 1000 states. As these states are not directly measurable, a state observer needs to be developed. In this paper, we compare three different state estimation techniques commonly applied to optimal control in buildings by applying them on a simulation model of the office building but fed with real measurement data. The considered observers are stationary Kalman Filter, time-varying Kalman Filter, and Moving Horizon Estimation. Summarizing the results, all estimators can achieve low output estimation error, but on the other hand only Moving Horizon Estimation is capable to keep the state trajectories within the limits thanks to the constraints at expenses of the computational time. As a first step towards real implementation of white-box MPC, in this paper, we have compared different state estimation techniques commonly applied to optimal control in buildings. We selected three different state observers available from the literature and compared their estimation error and robustness against initial conditions and noise in a numerical case study by using a virtual test bed model of a real building.