Computationally efficient, approximate moving horizon state estimation for nonlinear systems

Moving horizon estimation for discrete-time nonlinear systems is addressed by using fast optimization algorithms for which stability results under general conditions are ensured. The solution of the on-line moving horizon estimation problem is obtained by using the sampling time to solve a reference problem with model-predicted measurements while waiting for the next measurement. In order to correct the resulting solution, a quick nonlinear programming sensitivity calculation is accomplished as soon as the new measurement becomes available. The stability properties of such moving horizon estimation algorithm is proved under general conditions, which make the overall approach suitable for real settings with strong nonlinearities.