Statistical Learning for Optimal Control of Hybrid Systems

In this paper we explore a randomized alternative for the optimization of hybrid systems´ performance. The basic approach is to generate samples from the family of possible solutions, and to test them on the plant´s model to evaluate their performance. This result is obtained by first presenting the general hybrid optimal control problem, and then converting it into an optimization problem within a statistical learning framework. The results are applied to examples already existing in the literature, in order to highlight certain operational aspects of the proposed methods.