Authors: Puya Afshar, Hong Wang
Addresses: Control Systems Centre, The University of Manchester, P.O. Box 88, Manchester M60 1QD, UK. ' Control Systems Centre, The University of Manchester, P.O. Box 88, Manchester M60 1QD, UK
Abstract: The stochastic distribution control is important for certain industrial applications and cases where non-Gaussian noises exist. The problem has been initially solved by controlling the dynamical system formed by neural networks which approximate the output probability density function (PDF). Also, a modified version of iterative learning control (ILC) has been previously introduced by the authors to solve the output PDF shaping problem for linear weight dynamical systems. Looking into unknown non-linear weight dynamics, this paper presents a model reference neuro-adaptive control (MRNAC) approach for PDF shaping in non-Gaussian stochastic systems. The method is based on ILC and employs a neural network framework for modelling and controller. The time domain is first split up to batches. Then the proposed ILC method is implemented in two main domains namely within each batch and between any two adjacent batches. The design is carried out in three stages: a) NN-based non-linear dynamic system identification; b) MRNAC of the weight control loop within each batch; c) tuning the NN centres and widths between any two adjacent batches. Simulations confirm the effectiveness of the method.
Keywords: stochastic distribution control; SDC; iterative learning control; ILC; neural networks; model-reference adaptive control; MRAC.
International Journal of Advanced Mechatronic Systems, 2010 Vol.2 No.1/2, pp.108 - 116
Published online: 10 Jan 2010 *Full-text access for editors Access for subscribers Purchase this article Comment on this article