Setting up a framework for model predictive control with moving horizon state estimation using JModelica
Mats Vande Cavey, Roel De Coninck, Lieve Helsen · Linköping electronic conference proceedings · 2014
A model predictive control framework for optimal heating of a residential building is proposed.The control inputs are applied to a virtual building emulator model using a limited amount of measurements.State estimation is implemented using moving horizon estimation to reinitialize the states of the controller model in every time step.To implement the moving horizon estimation, the Modelica equations had to be modified.A stochastic input is declared at the controller model state equations to represent the process noise (model error).The state estimation significantly improves the output matching between emulator and controller model.The JModelica optimization framework proves to be satisfactory for this first, limited case investigated here.Future work will focus on the extension to different models and prediction errors within the framework developed here.