Characterization of Prediction Uncertainty using an Adaptive Fuzzy Rule based Technique

A.J. Abebe and R.K. Price (The Netherlands)


Fuzzy systems, prediction uncertainty, evolutionary algorithm, hydrodynamic models


This article presents a new approach to analyze and represent prediction uncertainty of models in the form of fuzzy IF-THEN rules. The prerequisites are the magnitudes of selected state variables while the consequence is the magnitude of the model prediction errors, both in linguistic and therefore human understandable form. A fuzzy rule-based system is used along with an evolutionary algorithm to extract rules from a historical data. The methodology is tested on a hydrodynamic model of a hypothetical estuary with intentionally introduced uncertainty. The results show that it is indeed possible to extract high level information from the performance of a model with historical data.

Important Links:

Go Back