Weather Time Series Forecasting Using Recurrent Fuzzy Neural Network

Thi Nguyen
Centre for GIS, School of Geography and Environmental Science, Monash University, Australia

     Full text: PDF
     Last modified: October 22, 2008
     Presentation date: 12/05/2008 3:45 PM in GIS2
     (View Schedule)

Abstract
Weather forecasting becomes more and more indispensable to our lives and thus many approaches have been investigated so far to meet this high demand. Conventional methods proposed in the 1980s or earlier were mostly linear models and usually applied to deal with short-range prediction. With the development of information technology, many data mining techniques have been introduced aiming to improve the power and accuracy in prediction. In this paper, three daily meteorological time-series encompassing maximum temperature, minimum temperature, and rainfall spanning from 1981 to 1990 in Melbourne, Australia were utilized to verify the prediction potential of the incorporation of fuzzy sets and neural networks by deployment of the Recurrent Fuzzy Neural Network (RFNN). Good experimental results were achieved via the application of the RFNN model to weather time series forecasting.