粒子群算法与支持向量机相结合的热带气旋强度预报试验

支持向量机(SVM)的惩罚参数及核参数的选择直接影响到模型效果,通过粒子群算法(PSO)解决支持向量机的参数选择问题,实现了参数选择的自动化。将该方法应用于热带气旋强度预报,利用气候持续性因子,挑选了1990年的100个左右样本进行预报检验,预报时效为12 h、24 h、36 h、48 h的强度平均绝对误差分别为3.00、4.35、4.93和6.68 m/s。另外,还与国外预报结果及采用最小二乘回归法的预报结果进行了效果的比较,SVM方法显示了更好的预报能力。

TROPICAL CYCLONE INTENSITY FORECAST EXPERIMENT BASE ON COMBINATION OF PARTICLE SWARM OPTIMIZATION AND SUPPORT VECTOR MACHINE

The model results of Support Vector Machine（SVM）are directly affected by the selection of regularization and kernel parameter.Particle Swarm Optimization（PSO） is introduced to select parameters for SVM,which achieves the automatization of parameter selection.The PSO-SVM model was used to Tropical Cyclone（TC）intensity forecasting,in which climatic persistence factors were utilized and about 100 samples in 1990 were chosen.The intensity mean absolute differences of 12 h,24 h,36 h and 48 h forecasting were 3.00 m/s,4.35 m/s,4.93 m/s and 6.68 m/s respectively.Furthermore,by making a comparative analysis of the foreign forecasting result and the least square method result,it showed that the SVM method had better capacity.