We present an extremum seeking control algorithm based on the estimation of the phasor of the perturbation frequency in the output of the plant. The phasor estimator is based on a continuous time Kalman filter, which is reduced into a variable gain observer by explicitly solving the special case of the Riccati equation. Local stability of the proposed al- gorithm for general non-linear dynamic systems using averaging and singular perturbations is presented for the single input case. The advantage of the presented algorithm is that it can be used on plants with large and even variable phase lag.
In this work we present a semi-global practical asymptotic stability analysis for phasor extremum seeking control with a general non-linear dynamic system. With the same technique applied to the classic band pass filter algorithm, we present a more relaxed (less constrained) semi-global practical asymptotic stability condition compared to earlier work. The results are based on a non approximated averaging for both control techniques.
Two desirable properties of a signal norm to be used for the purpose of optimal control or fault detection are highlighted. These properties are shift-invariance and the permitting of persistent signals. A class of norms, called the window norms, is analyzed. Among norms considered for control purposes, they appear to be the only ones, save the L∞-norm, that satisfy both properties. The signal spaces defined by the window norms, however, include some interesting signals that are not in the L∞-space. The window norms have been found suitable for application in fault detection and are here also considered for optimal control. It is shown that they can be taken as a support for the concept of L1-control but may also suggest a new class of optimal controllers.
We consider worst-case analysis of system identification under less restrictive assumptions on the noise than the l∞ bounded error condition. It is shown that the least-squares method has a robust convergence property in l2 identification, but lacks a corresponding property in l1 identification (as well as in all other non-Hilbert space settings). The latter result is in stark contrast with typical results in asymptotic stochastic analysis of the least-squares method. Furthermore, it is shown that the Khintchine inequality is useful in the analysis of least lp identification methods