Air temperature prediction using Support Vector Regression and GENIE: the Georgia Extreme-weather Neural-network Informed Expert
Chevalier, Robert Francis
MetadataShow full item record
Several studies have focused on comparisons between Support Vector Regression (SVR) and Artificial Neural Networks (ANNs). However, few have involved domains with massively large data sets. This research led to a methodology for reducing the number of SVR training patterns without a need to pre-process the data set. Using this methodology SVR models were created for air temperature prediction from one to twelve hours ahead. These models were more accurate than ANN models that were trained on data sets of 300,000 patterns and competitive with ANN models that were trained on 1.25 million patterns. A fuzzy expert system was also developed which incorporates the knowledge of local agrometeorologists in order to assess the risk of frost. Wind speed, as well as ANN models of air temperature and dew point temperature, enabled the expert system to make frost predictions from one to twelve hours ahead. This tool will be made available to Georgia farmers through a webbased interface that was created for The University of Georgia’s Automated Environmental Monitoring Network website (http://www.georgiaweather.net).