Research On Tropical Cyclone In The South China Sea Forecast Based On PSO-BP Neural Network

The tropical cyclone is one of the most destructive disasters in nature. The occurrence of tropical cyclone in Northwest Pacific is the most frequent and accounts for more than 1/3 of the global total number. So the study of tropical cyclone development and accurate forecasts of the tropical cyclone activity are of great significance for the prevention and mitigation of the disaster. Tropical cyclones developed in the South China Sea account for 36% of that in Northwest Pacific, so it is necessary to conduct further studies on tropical cyclones developed in the South China Sea and find an forecasting method with the relatively higher accuracy.With the development of science and technology, the further study of tropical cyclones and the enhancement of the means getting the data make the tropical cyclone data increase rapidly. So efficient means are required to manage, analyze, retrieve and share the tropical cyclone data. As the combination of GIS and Internet technology, WebGIS not only has the functions of GIS data management and spatial analysis, but also has the specialty of function of information publishing and data sharing, which provides a comprehensive platform for the tropical cyclone information publishing and data sharing.BP neural network with a strong nonlinear mapping ability has been applied in tropical cyclone forecasting, especially highly nonlinear tropical cyclone track forecasts. However, the parameters of BP neural network are based on partial information in the whole parameter space rather than the global optimal value, and thus it is easy to fall into local minimum, which will also reduce prediction accuracy of the BP neural network. Particle swarm optimization (PSO) with the function of global optimization is applied in this thesis to optimize the initial weights and thresholds of BP neural network, and thus creates a new PSO-BP neural network model for tropical cyclone forecasting.In this thesis, CMA-STI tropical cyclone track dataset is used as data resource. The PSO-BP neural network model is used for tropical cyclone track, intensity and pressure of 12h,24h,36h and 48h forecasts by using stepwise regression analysis to select the predictors. In comparison with PSO-BP neural network model, the BP neural network and CLIPER model, the experiment results indicate that the PSO-BP neural network has higher prediction accuracy, which proving that PSO-BP neural network model is feasible and practical in tropical cyclone forecasting.Finally, based on the historical tropical cyclone data and forecast data, this dissertation applies WebGIS technology to organize and manage data resources, uses ArcGIS API for Silverlight development package and builds a tropical cyclone data visualization system to visualize the tropical cyclone data. The system includes the following spatial analysis functions:tropical cyclone inquiry, the display of the historical tracks and the forecast tracks and the demonstration of the wind field and pressure field, which provides a good foundation for further study.