APPLICATION OF SVM NETWORK IN A HYBRID MODEL FOR WEATHER FORECASTING

Abstract

Weather forecast is a practical problem and have important implications for
agriculture, industry and other services. There have been different proposed
methods to forecast the weather parameters [3, 7, 8, 10], but the parameters of
the prediction model depends on the geographical conditions and the economic
development of the given area. Therefore, for every new location, we need to
find the parameters of the model or to propose a more suitable model. This
paper proposes to use the SVM network (Support Vector Machine) in a hybrid
model [2] to forecast the daily weather parameters (maximum temperature and
minimum temperature). The input data is the historical values of maximum and
minimum temperatures, humidity, wind speed and average values of rainfall,
sun hours for past days. Model inputs are evaluated and selected using linear
decomposition coefficients estimated using SVD (Singular Value Decomposition).
The quality of the proposed solution is tested on real environment data (taken
from 01/01/2010 to 31/12/2015, 2191 days) of Hai Duong province.