Research Of Classification Method For Tropical Cyclone Intensity Change Based On Evolutionary Algorithm

Forecasting for tropical cyclone (TC) landed intensity has been the challenging issue in the field of tropical cyclone regularity research. Because the change and development of TC is a complex natural phenomena, the accuracy of the forecasting for TC landed intensity is not high. There are a lot of complex parameter optimization problem in the field of TC intensity, especially the problems of the models and formulas. Evolutionary algorithm, which is proposed as a group search algorithm with direction, is very suitable for solving this kind of high-dimensional optimization problems. Therefore, with regard to the complex TC intensity parameter optimization problems, this paper advanced two kinds of TC intensity forecasting method based on evolutionary algorithm, and these works make a useful and necessary exploration for solving the more complex meteorological parameter optimization problems in future.This paper can be divided into the following three aspects in detail:1. This paper conducted an exploratory relative to the difficult and hot spot of the TC intensity forecasting problem, designing two kinds of models for TC intensity forecasting from the index model and classification rule angles. Besides, through in-depth studying of evolutionary algorithms and analyzing the current performance of most evolutionary algorithms, particle swarm optimization (PSO) and evolutionary strategy (ES) were selected in this paper. With regard to optimization to TC intensity forecasting model as a starting point, this paper took the advantages of group search in evolutionary algorithms to optimize the parameter of TC intensity models. Overall, the paper played out the advantages of interdisciplinary.2. This paper designed a prediction method of TC intensity based on the index model. In order to build the index model, the high-dimensional data was projected by projection pursuit (PP), and projection direction was optimized by ES. In the final, the index model was proposed to forecast the TC intensity. To verify the effectiveness of the method, using the actual dataset of TC, this paper compared final accuracy of C4.5 algorithm which used to the same TC intensity dataset. The experimental results showed the effectiveness and its own unique advantages of the method based on the index model, which providing a new way to the filed of TC intensity forecast.3. This paper also proposed a prediction method of TC intensity based on classification rules. From the perspective of the traditional classification algorithm, a set of classification rules mining process was designed. Hierarchical strategy, new coding scheme, data discretization method were combined with PSO algorithm, which used to mine the TC intensity classification rules. With regard to effectiveness of the method based on classification rules, using the same actual dataset of TC, this paper compared final accuracy with that of C4.5 algorithm and the method based on the index model. The experimental results demonstrated the effectiveness the method based on classification rules, which providing a feasible method for the rest of the classification for meteorology.The above work not only enriches the theory of forecasting for tropical cyclone intensity change, but also extends the application field of evolution algorithm. Evolution algorithm combined with the meteorology topics, it provides a new method and promotes the application of optimization in the meteorological research.