Nonlinear Time Series Prediction by Using RBF Network

Authors:
Liqiang Zhu
Affiliations:
School of Mechanical, Electronic and Control Engineering, Beijing Jiaotong University, Beijing, China 100044
Venue:
ISNN '09 Proceedings of the 6th International Symposium on Neural Networks on Advances in Neural Networks
Year:
2009

Citing 4
Cited 0

Comparison of algorithms calculating optimal embedding parameters for delay time coordinates

Conference proceedings on Interpretation of time series from nonlinear mechanical systems
Nonlinear time series analysis

Nonlinear time series analysis
Neural Networks: A Comprehensive Foundation

Neural Networks: A Comprehensive Foundation
A hybrid linear-neural model for time series forecasting

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper describes a numerical algorithm for short-term prediction of nonlinear time series by using time-delay embedding and radial basis function (RBF) neural networks. Unlike the existing RBF algorithms with centers preselected during training process and fixed during prediction process, the proposed method utilizes a simple selection algorithm to dynamically change the center positions, resulting in a local RBF model with time varying parameters. Analysis and methodology are detailed in the context of the Leuven competition. Results show that the proposed local dynamical RBF network performed remarkably well.