An improved neural network based targeting method for chaotic dynamics

In this study, an improvement in the previously proposed Extended Control Regions (ECR) method [Physica D 150 (2001) 163] is presented, which gives better performance with respect to training time and average reaching time while maintaining the advantages of the previous method. In the improved approach, the system data is first pre-processed and subdivided into clusters, and then one artificial neural network is assigned to each such cluster; thereby the short-time-predictability property of chaotic systems is exploited more efficiently. The performances of different ECR versions have been compared on basis of simulation results obtained for several chaotic systems. (C) 2002 Elsevier Science Ltd. All rights reserved.