Workshop on Optimal Control of Thermal Systems in Buildings using Modelica location:University of Freiburg date:23 - 24 March 2015
To guarantee indoor thermal comfort in a typical European office building, about 40% of the total average yearly primary energy use can be attributed to its HVAC systems. The energy use intensity of a building depends on many factors, amongst them: weather, building envelope, building use, etc.
When applying model predictive control (MPC), the control model is often a low order model, which is a simplified presentation of the real system capturing most of the relevant building dynamics. This model mismatch is usually taken into account when evaluating the controller performance. Another important -and often neglected- aspect related to real implementation is the bias on sensors and actuators. Both sensor and actuator bias can have a major impact on the achieved model based control (MBC, among them MPC) performance, especially when hard constraints are used. The impact of this bias can be evaluated through measurements (à posteriori) or through emulation.This contribution gives a simple example of the impact of sensor and actuator bias on a deterministic MPC controller (using the Modelica Library IDEAS), and discusses several methods of bias modelling in Modelica (manual coding, module, Dymola script). Also, the possibilities to include white noise or multiplicative errors on sensor and actuator signals during the controller model development are discussed.
This sheds a light on the robustness of the controller models towards sensor and actuator bias. These insights are critical when the (low order) models are used for evaluating in-use performance, or when they are to be included in a controller implemented in real installations.