Priori Information Based Support Vector Regression and Its Applications

Abstract

In order to extract the priori information (PI) provided by real monitored values of peak particle velocity (PPV) and increase the prediction accuracy of PPV, PI based support vector regression (SVR) is established. Firstly, to extract the PI provided by monitored data from the aspect of mathematics, the probability density of PPV is estimated withε-SVR. Secondly, in order to make full use of the PI about fluctuation of PPV between the maximal value and the minimal value in a certain period of time, probability density estimated withε-SVR is incorporated into training data, and then the dimensionality of training data is increased. Thirdly, using the training data with a higher dimension, a method of predicting PPV called PI-ε-SVR is proposed. Finally, with the collected values of PPV induced by underwater blasting at Dajin Island in Taishan nuclear power station in China, contrastive experiments are made to show the effectiveness of the proposed method.