Conference article

We introduce a tool for obtaining optimal control and estimation problems from graphical models. Graphical models are constructed by combining blocks that can be implemented in Modelica or taken from a palette. The models can be used for predictive control; moving horizon estimation; and/or parameter estimation. We show that the solution time and robustness of the resulting nonlinear program strongly depends on the way the model was built and translated. These observations yield modeling guidelines for increasing robustness and efficiency of the optimization. In particular; we find out that eliminating as many variables as possible from the optimization problem may help a lot.

Modelica; Optimization; Optimal Control; State Estimation; Receding Horizon; MPC; MHE