Reinforcement learning control using interconnected learning automata

Abstract

This paper is concerned with the application of interconnected learning automata to the control of unknown stochastic dynamic systems. A method of using N learning automata and subsets of actions to search for optimum controller parameters according to a given performance index is presented. The methodology has been applied as an on-line control strategy for control of an unknown turbogenerator system in noise environments without persistent excitation signals. Simulation results are included.