Solar flare forecasting model of combining principal component analysis with support vector machine

Abstract

Solar flare is an important event in the solar atmosphere. It is a main disturbance resource of space environment and has a dramatic impact on human activities. How to promote the forecast ability of solar flare has a great significance for human beings. To better predict solar flare, this paper used principal component analysis and support vector machine, which are two classic machine learning methods, to predict solar flare. In feature extraction part, principal component analysis is a generally adopted method, and it is a technique for linearly compressing multidimensional data into lower dimensions with minimal loss of information. In flare classification part, support vector machine is a learning machine based on the statistical learning theory. The support vector machine can deal with nonlinear problems in classification and regression easily by using kernel functions, which is necessary to map onto another high dimension linear space. In solar flare forecasting research, the relationship between solar flare and morphological evolution of sunspots plays an important role in daily flare forecasting. And there is evidence that sunspot parameters and 10.7 cm solar radio flux are extremely related with solar flare. With the predictors fully taken into account, a new method of combing principal component analysis with support vector machine is proposed to improve the forecast ability of solar flare. In this paper, the sunspot parameters are area of sunspot group, McIntosh classification, extended longitude, the sunspot number in the solar active region and magnetic classification. Using attribute coding and appropriate transform function, the initial data set of predictors is normalized to form the modeling data set. And based on this data set, principal component analysis and support vector machine are applied to build solar flare forecasting model. In experiment, the forecasting model is compared with other model, which works well in the solar flare short-term forecasting. This shows that the PCA-SVM prediction model has higher precision compared with the other one when either considering positive examples or negative samples. The results indicate that the Solar flare forecasting model of combining principal component analysis with support vector machine is an effective flare prediction model.