Analysis of Support Vector Regression Model for Micrometeorological Data Prediction

Abstract

This paper aims to reveal the appropriate amount of training data for accurately and quickly building a support vector regression (SVR) model for micrometeorological data prediction. SVR is derived from statistical learning theory and can be used to predict a quantity in the future based on training that uses past data. Although SVR is superior to traditional learning algorithms such as the artificial neural network (ANN), it is difficult to choose the most suitable amount of training data to build the appropriate SVR model for micrometeorological data prediction. The challenge of this paper is to reveal the periodic characteristics of micrometeorological data in Japan and determine the appropriate amount of training data to build the SVR model. By selecting the appropriate amount of training data, it is possible to improve both prediction accuracy and calculation time. When predicting air temperature in Sapporo, the prediction error was reduced by 0.1°C and the calculation time was reduced by 98.7% using the appropriate amount of training data.