23. A Study on Typhoon Risk Prediction by Different Methods of Pattern Recognition

This chapter presents the economic loss prediction of typhoon by different pattern recognition methods, such as multivariable statistics (MS), case base reasoning (CBR), fuzzy theory (FT), and neural network model (NN). The typhoon records in Taiwan before 2000 were used as the database for reference, and the records after the year 2000 were predicted using the pattern derived from the database. Six scenarios were calculated using these methods. The first sceneries include the parameters: maximum wind speed, minimum atmospheric pressure, maximum wind speed in typhoon center and lowest atmospheric pressure near typhoon center. The second scenario includes the previous four parameters with rainfall and calculated by CBR. The third scenario uses the fuzzy calculation with five parameters. The successful rate of prediction for the three methods was 12.5, 37.5, and 57 %. The results reveal that the fuzzy calculation can significantly increase the prediction rate than the traditional CBR method. On the other hand, five neural network methods were compared, which were back propagation network (BPN), extend neuron networks (ENN), fuzzy neural network (FNN), analysis adjustment synthesis network (AASN), and genetic algorithm neural network (GANN). The result reveals that the BPN is the best choice, because the error is the lowest among the five schemes in this study.