MPC Performance Assessment and Nonlinear State Estimation
Joining the group in 2009, Luo Ji’s research focuses on controller performance evaluation and state estimation of nonlinear systems.
MPC performance evaluation is an important issue in theoretical implementation and industrial application. Due to the existence of white noise (which may contain process and measurement noises), unmodeled disturbances, noise model mismatch, or even process model mismatch, the output performance and input cost may be quite different even if the same control strategy is used. To address this issue, we have proposed a stage cost function and examined its value by lab simulations. This could be applied as the evaluation criterion between different controller results, even for some which are not MPC controllers. The cases when unmodeled disturbances or model mismatch exist are also studied.
The evaluation and implementation of nonlinear state estimators is also critical in real applications since (1) most industrial systems have many states which cannot be measured directly and (2) the controller needs predictions of these states to make the right decision. From previous work, Moving Horizon Estimators (MHE) could be a good candidate method to depict the state space from nonlinear processes. My work includes (1) theoretically designing and verifying a stable, convergent, well-defined nonlinear MHE on realistic industrial plants, (2) practical implementation of nonlinear MHE algorithms on real chemical plants, and (3) more advanced and simplified estimation algorithms for special operating conditions.