Direct current motor control based on high order neural networks using stochastic estimation

Abstract

An adaptive discrete-time tracking controller for a direct current (DC) motor with controlled excitation flux is presented. A high order neural network in discrete-time is used to identify the plant model; this network is trained with an extended Kalman filter where the associated state and measurement noises discrete-time covariance matrices are calculated with stochastic estimation. Then, the discrete-time block control and sliding mode techniques are used to develop the trajectory tracking for the angular position of a DC motor with separate winding excitation. Numerical computation presented in this paper shows that the proposed method provides accurate estimation for the covariance matrices associated in the extended Kalman filter.

Real-Time Control Based On High Order Neural Networks Using Stochastic Estimation

“…where x i , i = 1, · · · , n is the state of the ith neuron, χ(k) denotes the plant state, n is the state dimension, w i , i = 1, · · · , n is the respective on-line adapted weight vector, and z i (χ(k), u(k)) is given as in [1]. This RHONN is trained with the EKF algorithm.…”

“…The training goal is to find the optimal weight values which minimize the prediction error. In this paper, we use the EKF described as in [1].…”

“…The proposed stochastic formulation requires the efficient compute of Q i (k) and R i (k) in recursive form such that minimizes the identification error [1] …”

“…On the other hand, artificial neural networks (NN) have become an attractive tool which can be used to construct a model of complex nonlinear processes ( [4], [9], [18]); particularly, the used of discrete-time high order neural networks for identification and control has recently increased ( [2], [4]) where the extended Kalman filter (EKF) based algorithms have been introduced to train these NN and to improve the learning convergence ( [1], [7], [19], [20]), among others.…”

“…Among those, block control (BC) combined with sliding mode (SM) constitutes a well suited design methodology ( [1], [11]). Nevertheless, as well as several feedback linearization schemes the BC technique requires non-singularity of some parameters in order to preserve controllability of NN.…”