Artificial neural networks for stochastic control of nonliner state space systems

Abstract

In this paper, stochastic control of nonlinear state space models is discussed. After a brief review on nonlinear state space models, a multi layer perceptron (MLP) neural network is considered to represent the general structure of the controller. Then, an expectation maximization (EM) algorithm joint with the particle smoothing framework are proposed for updating parameters of the MLP network. The suggested structure is also applied to the trajectory tracking of a nonlinear/non-stationary system. Simulation results show the superiority of our method in the control of nonlinear and stochastic state space models.