姬战生 章国稳 黄薇

摘要 针对平原河网地区河道水位预报的不确定性特征，以京杭运河代表站拱宸桥站为例，提出了一种基于在线支持向量机的河道水位预报方法。根据拱宸桥站河道水位影响因子，选取京杭运河2008—2013年11场典型洪水过程的1 500组水文数据作为样本，分别构建了固定式、在线增量式和在线剔除式SVM水位预测模型。定性定量比较了不同预见期下3种模型的预测数据，结果表明在线剔除式SVM模型的预报精度要高于其他2种模型。该方法可为研究平原河网地区河道水位过程实时预报提供参考。

关键词 支持向量机;在线剔除式;平原河网地区;水位过程预报

中图分类号 TV 124  文獻标识码 A  文章编号 0517-6611（2021）14-0191-05

Abstract Due to the uncertainty characteristics of forecasting river water level of plain river network， this paper took the representative station Gongchenqiao Station of Jinghang Canal as an examples and proposed a forecasting method of river water level based on online support vector machines（SVM）. According to the influencing factors of the water level at Gongchenqiao Station， 1 500 sets of hydrological data from 11 typical flood processes on the Jinghang Canal from 2008 to 2013 were selected as samples and fixed， online incremental and online elimination SVM water level prediction models were constructed. The forecast data of the three models under different forecast periods were qualitatively and quantitatively compared. The results showed that the prediction accuracy of the online elimination SVM model was higher than that of the other two models. This method could provide references for studying the realtime prediction of the water level process in the plain river network.

Key words Support vector machines;Online elimination;Plain river network regions;Water level process forecasting

基金项目 国家自然科学基金项目（51705114）;浙江省自然科学基金项目（LQ16E080009）;浙江省教育厅一般科研资助项目（Y201430581）;杭州市科技发展计划项目（20191203B72）;浙江省水利科技计划项目（RC1807，RC1901）。

作者简介 姬战生（1980—），男，河南洛阳人，高级工程师，硕士，从事水文预报和钱塘江涌潮预报研究。

收稿日期 2020-10-28

平原河网地区多为经济发达地区，城市化速度较快，导致降水汇流时间减少、洪峰出现时间提早、洪峰流量变大，使原有的河道防洪能力已不能满足要求，洪涝灾害问题愈加突出[1]。水文预报是流域防洪减灾的基础，是防洪决策的主要依据，及时、准确、可靠的洪水预报可对平原河网地区防洪排涝带来巨大的经济效益和社会效益。笔者探究适用于平原河网地区的河道水位过程实时预报方法，为区域防灾减灾提供科学决策依据，促进经济社会的健康发展。……