Summary
Extremum-seeking control is a purely data-based performance optimization strategy for dynamical systems. Typical extremum seeking approaches rely on applying perturbations to tunable system parameters and measuring the corresponding system output, to determine a suitable search direction aimed at optimizing steady-state performance. While in practical applications performing measurements may be costly or time-consuming, these measurement data are typically discarded after a single parameter update step. To make more efficient use of these valuable data, we propose a novel extremum-seeking control approach that uses the data to construct an online approximation of the system’s steady-state performance map, instead of discarding them. By using this approximation to determine the search direction and update step size whenever the approximation is deemed sufficiently accurate, the total number of measurements needed to optimize performance is significantly reduced, as shown in an illustrative example.