Forecasting of Chaotic Time Series Using RBF
Neural Networks Optimized By Genetic Algorithms
Mohammed Awad
Faculty of Engineering and
Information Technology, Arab American University, Palestine
Abstract: Time series forecasting is an
important tool, which is used to support the areas of planning for both
individual and organizational decisions. This problem consists of forecasting
future data based on past and/or present data. This paper deals with the
problem of time series forecasting from a given set of input/output data. We
present a hybrid approach for time series forecasting using Radial Basis
Functions Neural Network (RBFNs) and Genetic Algorithms (GAs). GAs technique
proposed to optimize centers c and width r of RBFN, the weights w of RBFNs
optimized used traditional algorithm. This method uses an adaptive process of
optimizing the RBFN parameters depending on GAs, which improve the homogenize
during the process. This proposed hybrid approach improves the forecasting
performance of the time series. The performance of the proposed method
evaluated on examples of short-term mackey-glass time series. The results show
that forecasting by RBFNs parameters is optimized using GAs to achieve better
root mean square error than algorithms that optimize RBFNs parameters found by
traditional algorithms.
Keywords: Time series forecasting, RBF neural networks, genetic algorithms, hybrid
approach.
Received March 17, 2015;
accepted October 7, 2015