Our modified back-propagation algorithm for FMP can adapt chaotic time series very well. Together with the parallelization of batch learning under PVM the enormous amount of trainig time can be significantly reduced.
For the future we will adapt our implementations to the prediction of time series for economical consum processes.