Introducing an Incremental Learning Method for Neuro-fuzzy Models with the Application to Forecast Natural Chaotic Dynamics

Predicting jiiture behavior of chaotic time series and systems is a challenging area in nonlinear prediction. The prediction uccurucy of chaotic time series is extremely dependent on the model and on the learning algorithm. In addition, the generalization property of the proposed odels trained by limited observations is of great importance. In this study, the recently developed neuro-ftzzyin terpretation of locally linear models, which have led to the introduction of intuitive incremental learning algorithms e.g. LoLiMoT, are implemented in their optima1 structure to be compared with several other methods in forecasting natural chaotic dynamics. The scope of paper is to reveal the advantages of neuro-$tizzy models in comparison with the most successful neural and ji~zzy approaches in their best structures in predicting chaotic dynamics according to prediction accuracy, generalization, and computational complexity. The Muckey-Glass chaotic time series us a benchmark and Sunspot number and Darwin sea level pressures time series are considered as practical examples of chaotic time series

Introducing an Incremental Learning Method for Neuro-fuzzy Models with the Application to Forecast Natural Chaotic Dynamics Keywords:

forecasting , chaotic time series , locally linear newo-fuzzy model , LoLiMoT