基于3种机器学习算法的台风频数预测  

Typhoon number prediction based on three machine learning algorithms

作　　者：荣新 覃卫坚[2] 韦文山[1] RONG Xin;QIN Weijian;WEI Wenshan(School of Electronic Information,Guangxi Minzu University,Nanning 530006,China;Guangxi Climate Center,Nanning 530022,China)

机构地区：[1]广西民族大学电子信息学院,广西南宁530000 [2]广西气候中心,广西南宁530022

出　　处：《海洋预报》2023年第5期1-9,共9页Marine Forecasts

基　　金：广西科技计划项目(桂科AB21075005);广西自然科学基金(2019GXNSFAA245048)。

摘　　要：为了提高影响广西台风频数的年度预测准确率,利用中国气象局上海台风研究所提供的1951-2020年影响广西的台风样本数据、国家气候中心提供的88项大气环流特征量和26项海温指数资料,使用相关方法找出高影响因子。针对影响台风物理因素的复杂性,为了获取更综合的预测因子信息,使用随机森林对影响因子进行二次筛选,建立基于随机森林、支持向量回归和循环门单元(GRU)3种机器学习算法的影响广西台风频数气候预测模型。实验结果表明:使用随机森林二次筛选得到的因子的建模预测效果明显提高,机器学习算法预测效果整体高于岭回归方法,其中GRU预测效果最好,绝对误差较岭回归方法减少10.30%,其次为随机森林和支持向量回归,误差分别减少9.44%和7.47%。In order to improve the prediction accuracy of annual number of typhoons affecting Guangxi,this paper uses related methods to find high impact factor based on the sample data of typhoons affecting Guangxi from 1951 to 2020 provided by Shanghai Typhoon Institute of China Meteorological Administration,the 88 atmospheric circulation feature quantities and 26 SST index data provided by the National Climate Center.In view of the complexity of physical factors in typhoon number forecasting,in order to obtain more comprehensive factor information,the random forest is used to screen the factors,and a prediction model for annual number of typhoons affecting Guangxi utilizing three machine learning algorithms,i.e.Random Forest,Support Vector Regression and Gate Recurrent Unit(GRU),is established.The results show that the prediction ability of using factors selected by Random Forest screening is significantly improved,and the prediction ability of using machine learning algorithms is higher than that of Ridge Regression method.Among them,GRU has the best prediction,and the absolute error is reduced by 10.30%compared with Ridge Regression method,followed by Random Forest and Support Vector Regression,with errors reduced by 9.44%and 7.47%,respectively.

关 键 词：影响广西台风频数 特征选择 随机森林 支持向量回归 循环门单元 

分 类 号：P444[天文地球—大气科学及气象学]