Moving horizon estimation is an estimation technique, which is utilized to optimally estimate the state of systems with constraints and/or nonlinearities. It utilizes previous inputs and measured outputs to estimate the current state and possibly predict the future evolution of the system. In contrast to e.g. the extended Kalman filter MHE has the advantage that it does not only use the last output, but also older outputs. Moreover, it can handle constraints directly. This allows an improved estimation performance. However since the MHE requires at each time instance the solution of an optimization problems it has a higher computational load compared to e.g. the EKF, which can be prohibitive for the application of MHE to real world problems.
The general aim of this work is to investigate different techniques, which can help to reduce this effort. More precisly, there are different possibilities to tailor the MHE to the problem: Often the underlying problem is solved using by a combination of discretization and nonlinear optimization. Possible working packages for this project include the investigation/comparison of different optimization techniques, problem formulations or discretization schemes.
If you are interested in this topic, please contact me.Literature search/review: | ca 30% | |
Systems theory: | ca 20% | |
Implementation/Test: | ca 50% |