Title: Volterra series based nonlinear system identification methods and modelling capabilities

Authors: Gargi Trivedi; Tarun Kumar Rawat

Addresses: Department of Electronics and Communication Engineering, Netaji Subhas University of Technology, New Delhi, India ' Department of Electronics and Communication Engineering, Netaji Subhas University of Technology, New Delhi, India

Abstract: The current study discusses Volterra series based nonlinear system models such as the Taylor series, time-delay neural network (TDNN) and nonlinear autoregressive model (NARX). The study aims to construct a truncated second-order Volterra model that can be used to identify nonlinear systems and compare their performance to that of the TDNN using benchmark cases. The feasibility of the feedback and feedforward networks is evaluated using a dataset of cortical responses evoked by wrist joint manipulation. It is observed that the TDNN is a mathematical model with more customisable parameters and requires less computation time than the Volterra system with particle swarm optimisation (PSO). Also, open-loop connections with fewer a-prior system assumptions, such as the Volterra model, can estimate 42% of wrist dynamics and closed-loop connections like the NARX model can estimate 93% of complex nonlinear dynamics.

Keywords: Volterra; time-delay neural network; TDNN; nonlinear autoregressive model; NARX; model structure; particle swarm optimisation; PSO.

DOI: 10.1504/IJMIC.2022.127513

International Journal of Modelling, Identification and Control, 2022 Vol.41 No.3, pp.222 - 230

Received: 15 Aug 2021
Accepted: 06 Dec 2021
Published online: 07 Dec 2022 *

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