N. Sarmadi and M. Teshnehlab (Iran)
Forecasting, Neuro-Fuzzy network, Time series, Information fusion, Emotional learning, and chaotic system
Weather Forecasting has been one of the most challenging problems around the world for more than half a century. On the other hand, Intelligent modeling has gained a lot of attention and interest recently . In this paper, a Neuro-Fuzzy (NF) method for forecasting of three major meteorological parameters (Pressure, Temperature and Relative humidity) is proposed. The availability of historical weather data makes this area highly suitable for Artificial Neural Networks (ANN’s) implementation. ANN’s are able to learn (extract) the relationships between past and current so predict future climatic variable values, combining both time series and regression approaches. To obtain desired system that will good approximate (generalize) and forecast (interpolate and extrapolate) the future values of mentioned atmospheric (chaotic system) parameters, different weights (degrees of freedom) in different layers are adjusted (tuned) either by Back-Propagation (BP) or Emotional Learning (EL) algorithms. Also spatial or temporal information fusion is used to reach better results (from multiple point sources). The power and prediction ability of model is evaluated by graphical (qualitative) and some quantitative criteria (performance functions) like mean absolute error, root mean square of error and correlation coefficient which show that proposed structure is an excellent tool for our goal.