Title: Adaptive deep neural network tracking controller augmented using only one-neuron hidden layers for nonlinear systems subject to high constraints and unknown uncertainties

Authors: Hampu Ait Abbas

Addresses: Laboratory of Materials and Durable Development (LM2D), Akli Mohand Oulhadj University – Bouira, Drissi Yahia Street, 10000, Bouira, Algeria

Abstract: A new control strategy, adaptive tracking controller augmented using deep learning hybrid method, is proposed to achieve excellent tracking performances of nonlinear systems in the presence of structured and unstructured uncertainties. Since conventional controllers suffer from limitations due to the presence of these uncertainties, we contribute in this paper to demonstrate the feasibility of applying deep learning (DL) algorithm as an approximator for neglected dynamics and uncertain parameters. Thus, the key idea of the developed adaptive FLC augmented using DNN is to both replace the conventional controller Dcom and compensate adaptively the effect of modelling errors for Highly Uncertain NLSs. The weight adaptation rule of the DNN is derived from the Lyapunov stability analysis that ensures boundedness of the error signals. Simulations of the proposed adaptive controller based DNN are conducted then compared to the Dcom without DNN, PI controller, and adaptive controller based SHLNN to demonstrate its practical potential.

Keywords: nonlinear systems; NLSs; feedback linearisation control; FLC; dynamic compensator; Dcom; single-hidden-layer neural network; SHLNN; deep learning; deep neural network; DNN; supervised fine-tuning; SFT process; layer-wise non-supervised learning algorithm; LWLA; backpropagation.

DOI: 10.1504/IJMIC.2021.121839

International Journal of Modelling, Identification and Control, 2021 Vol.37 No.3/4, pp.331 - 343

Received: 10 Aug 2020
Accepted: 08 Feb 2021

Published online: 07 Apr 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article