Optimization of Artificial Neural Network Architectures for Time Series Prediction Using Parallel Genetic Algorithms

Abstract

This paper considers the application of parallel genetic algorithms to the optimization of modular neural network architectures for time series prediction. We have a cluster configuration of 16 computers and the application is executed using the Matlab Distributed Computing Engine included in MATLAB r2006b. The Linux Fedora Core VI Operating System was installed and configured for the cluster execution due to its high performance, scalability and because it presents innumerable benefits that facilitate the implementation of distributed computing applications. The first part of this paper presents the theoretical framework with basic concepts like times series, artificial neural networks, genetic algorithms, and parallel genetic algorithms. The second part of this paper presents the procedure for configuring the cluster of computers, requirements, experiences and main problems that were encountered. Also, the development of the project is presented explaining as it was initially proposed and the adjustments that were required. The third part of this paper presents the obtained results for the time series prediction using tables, graphics and describing each one of them. Finally the conclusions and future works are presented.KeywordsGenetic AlgorithmHide LayerSoft ComputingNeural Network ArchitectureTime Series PredictionThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.