A prediction method of spatiotemporal series based on support vector regression model

Abstract

In this paper, a series of spatiotemporal data is analyzed by a regression method based on the theory of support vector machine (SVM). The support vector regression (SVR) model is used to predict the remote sensing data sets effectively. Firstly, we studied how to build a SVR model for spatiotemporal series prediction, and studied the problems of the test data processing, model parameters selection and kernel function construction. Secondly, the kernel functions used in previous studies were extended, and a new method for constructing spatiotemporal kernel function is proposed by using the mixed kernel function through comparative analysis for different kernel functions and model parameters. Finally, the obtained model is tested by using remote sensing evaluation data of eco-environmental vulnerability. The predicted results were compared with that obtained by using other classic kernel functions. It shows that the model proposed in this paper is more accurate to other classical models. Meanwhile it also can be found that the data of longer the time range is calculated, the better accuracy the prediction effect.