A Genetic-Neural Network Model Based on Multidimensional Scaling for Typhoon Intensity

Basing on the sample of typhoon from 2001 to 2010 for 10 years in the Northwest Pacific (NP), setting up the genetic-neural network prediction (GNNP) model which input predictors is using the methods of multidimensional scaling analysis (MDS) and Stepwise regression basing the predictors of climatology persistance to predict the typhoon intensity for 12, 24, 36, 48, 60 and 72 hour. The experimental forecast results showed that the average absolute forecast error of 30 independent samples of typhoon intensity in the Northwest Pacific 12-72h by the new model is 3.83, 4.72, 5.20, 6.44, 6.48 and 6.48m/s, respectively. Moreover, comparison the results of the new model and the Stepwise regression model under the condition of the same typhoon samples and the same forecast factors, the consequence indicates that the genetic-neural network prediction model which basing on the MDS is obviously more skillful than the Stepwise regression model. Apart from the forecast errors of 12h which is correspond of the result by Stepwise regression model, other average absolute error respectively fell 0.54, 1.1, 0.65, 1.09 and 2.12m/s.