摘要
风暴潮数值模型预报结果在小尺度区域往往存在弱于或强于实况的问题。针对此问题,该风暴潮灾害模拟方法融合数值模型预报结果为特征要素,使用一种基于多要素MOS(Model Output Statistics)风暴潮灾害过程模拟方法来实现对风暴潮灾害的预报。该模拟方法是将低维的特征向量映射到高维的隐含层中进行支持向量机(SVM)学习训练并控制输出相对误差,其特点在于MOS风暴潮灾害过程模拟中融入数值模型预报结果,并通过在一定的约束条件下最小化不敏感损失函数的逐渐优化逼近实测值来实现。实例验证结果表明:在风暴潮灾害过程减增水切换时期,数值模型预报的水位增水峰值为0.95 m,低于水位实际增水峰值1.1 m,历时5 h,预报曲线表现较为平滑,而MOS风暴潮模型在此期间模拟得到历时2 h就达到水位增水峰值1.14 m,模拟曲线爬升明显较快,这与实际风暴潮灾害过程的特征更加吻合;从预报结果的均方误差(RMSS值)和相关系数(CORR值)来看,MOS风暴潮模型预报结果的RMSE值和CORR值分别为0.165 m和0.945,相比数值模型的0.190 m和0.912都有所提高,可为后期精细化风暴潮预报工作提供一种新的过程模拟思路。
The storm surge numerical model predicts that the water increment process is weaker or stronger than the actual situation.This paper conducts the statistical analysis on the historical data of storm surge,and fuses a characteristic factors which are the result of numerical model into the storm surge disaster process simulation by using the MOS(Model Output Statistics) storm surge model.With this method,the low-dimensional feature vectors are mapped to the high-dimensional hidden layer to control the relative error with Support Vector Machine(SVM) learning,in which the minimized insensitive loss function approximates the measured data under certain constraint condition.The results show that during the switching period of decreasing and increasing water in the storm surge disaster process,the peak value of water increment is 0.95 m below the actual peak value of 1.1 m for 5 hours,and the MOS storm surge model figures out that the peak value of water increment climbs to 1.14 m within 2 hours,which is more consistent with the actual features of water increment process.The reason may be that the numerical model incorporate into the MOS storm surge model,and the measured values are approximated by the gradual optimization of the insensitive loss function under certain constraints.The the root mean square error(RMSE) and the correlation coefficient(CORR) of the forecast results show that RMSE and CORR of MOS storm surge model are 0.165 m and 0.945 respectively,which are higher than those of the numerical models of 0.190 m and 0.912.The paper provides a new idea for the refined storm surge forecasting operation in the later stage.
作者
张广平
彭世球
张晨晓
ZHANG Guangping;PENG Shiqiu;ZHANG Chenxiao(Ocean College ,Beibu Gulf University ,Qinzhou 535011,China;Guangxi Key Laboratory of Marine Disaster in the Beibu Gulf,Qinzhou 535011, China;College of Food Engineering, Beibu Gulf University,Qinzhou 535011,China)
出处
《安全与环境工程》
CAS
北大核心
2019年第3期50-55,共6页
Safety and Environmental Engineering
基金
广西自然科学基金重点项目(桂科AB18294017)
广西自然科学基金项目(2015GXNSFAA139242)
广西高校中青年教师基础能力提升项目(KY2016YB477)
钦州市科学研究与技术开发计划项目(20164411)
热带海洋环境国家重点实验室(中科院南海海洋研究所)开发课题项目(LTO1507)
广西北部湾海岸科学与工程实验室课题项目(2016ZYB14)
广西北部湾海洋灾害研究重点实验室课题项目(2019TS06)
关键词
风暴潮
MOS模型
灾害过程模拟
特征要素
支持向量机(SVM)
海洋灾害
storm surge
MOS (Model Output Statistics) model
disaster process simulation
characteristic factors
Support Vector Machine(SVM)
ocean disaster
