B-Spline neural network and chaotic harmony search applied to yo-yo motion system identification

Abstract

In an attempt to increase the accuracy in modelling nonlinear systems, a wide variety of mathematical models were developed, such as the Volterra series, Wiener models, bilinear models, Hammerstein-Wiener models, among others. Recently, artificial neural networks (ANNs) have generated considerable interest as alternative nonlinear system identification technique and regression tool. Unlike other models, ANNs are easy to adapt and to apply to different engineering problems. One example is the B-spline neural network (BSNN), which consists in a variation of the basis function artificial neural network, which is usually trained by gradient-based methods. However, gradient-based methods have the characteristic of frequently fall into local minimum during learning procedures. To overcome this problem found in conventional learning methods, Harmony Search (HS), a stochastic metaheuristic approach, can be adopted. This optimization technique was originally conceptualized using musical improvisation process of searching for a perfect state of harmony, and can be considered to adjust the control parameters of BSNNs. In the same way, this paper proposes uses a Chaotic Harmony Search (CHS) approach, based on Chaotic Logistic Map, to train BSNNs to reproduce the dynamic of a nonlinear yo-yo system. Numerical results indicate that the proposed CHS approach is effective in building a reasonable BSNN model for nonlinear identification.