Chaotic time-series prediction and the Relocating-LMS (RLMS) algorithm for radial basis function networks

In this study, the problem of real-time chaotic time-series prediction using Radial Basis Function Networks is addressed. The performance of a number of training methods based either on supervised error correction or on adaptive clustering techniques are investigated. Some performance drawbacks due to their exclusive usage are pointed out and a new algorithm combining their desirable properties is presented. The proposed Relocating-LMS algorithm is compared with the existing methods on a chaotic time-series produced by the Mackey-Glass Equation and is further tested on a series generated by the Logistic Map function, leading to encouraging results.