Study on Index Model of Tropical Cyclone Intensity Change Based on Projection Pursuit and Evolution Strategy

Abstract

This paper deals with the forecasting of tropical cyclone (TC) landed intensity change problem in which multi-level and multi-attribute decision are considered. A novel index model of tropical cyclone intensity change based on projection pursuit (PP) and evolution strategy (ES) is proposed to forecast the TC intensity change. We propose to use projection pursuit to project the high-dimensional TC intensity observation samples with 18 different attributes into one-dimensional projection index value. According to the projection index value distribution of learning samples, including TC intensifying and weakening, we can determine the cut off index value which distinguishes two different index value of intensifying and weakening samples. The final best projection unit vector can reflect degree of each attribute influence on TC intensity change. In order to solve the high-dimensional globally optimal problem in PP, evolution strategy with stochastic ranking is used to optimize the projection vector. The role of stochastic ranking is to balance the dominance of objective and penalty functions. Based on the index model, experimental results indicate that the accuracy of 693 TC intensity change samples reaches 89.2% when we take the index value 1.40 as the cut off value between TC intensifying and TC weakening, and the seven core attributes can also reflect the main meteorological characters of TC intensity change accurately.