Regresar

Learning an Improved LMI Controller Based on Takagi-Sugeno Models via Value Iteration

Abstract:

This article proposes an alternative for formulating the method to improve the conservative controllers based on linear matrix inequality (LMI), action-value function (Q-function), and value iteration algorithm to learn optimal controllers by using system data. In this respect, the proposed uses ideas of the previous works that parametrize in a particular way the Q-function. In this sense, the Q-function can be described with polynomials membership functions for fuzzy models of Takagi-Sugeno and initialize a learning process with the LMI controller. The obtained controller uses both the information about the membership functions and a set of data obtained from the system to improve the LMI controller. A TORA system is used to illustrate the approach.