ABSTRACT
The linear and nonlinear dimensionality reductions of the 21 primary selection factors for the data of strong and super typhoons that landed in South China in the period of 1990–2016 are extracted through Alasso and combined weight method, and the linear and nonlinear characteristics of the disaster system are extracted in the low-dimensional subspace. The support vector machine (SVM) parameters are optimized, and an assessment model that combines the drag onfly algorithm and SVM (DA-SVM) for extreme disaster losses caused by typhoons is constructed through the dragonfly group optimization algorithm. The results show that the average absolute and relative errors of the proposed new nonlinear assessment model are 22.993 and 10.65%, respectively, and the root mean square error is 0.0315. A comparison of this evaluation result with those of the traditional SVM regression model calculation shows that the stability, fitting effect, and evaluation accuracy of the DA-SVM model based on Alasso and combined weights are significantly improved compared with the conventional SVM regression method in estimating the extreme disaster losses of typhoons in South China.
