在年径流预报中, 气象因子之间的自相关会对预报精度产生影响。针对这个问题, 将主成分分析( PCA) 和粒 子群优化( PSO) 算法加入 SVR 模型中, 建立了 PCA2PSO2SVR 预报模型, 剔除冗余信息和噪声, 提取因子间的主要 特征, 并选择模型的最优参数组合作为回归支持向量机( SVR) 模型的输入。选择南水北调中线水源地丹江口水库 为研究区, 利用丹江口 1981- 2016 年入库资料进行模型检验。结果表明, 模型验证期间合格率为 831 33% 、距平一 致率也达到 831 33 % , 具有精度高稳定性强等优点, 对丹江口水库年径流预报有一定的参考意义。

In the annual runoff fo recast, the auto correlat ion betw een meteo ro log ical facto rs w ill have an impact o n fo recast accu2 r acy. In o rder to so lv e this problem, w e added PCA and PSO to the SVR model and established a PCA2PSO2SVR model. We re2 mo ved the r edundant informat ion and no ise, ext racted the main features of the facto rs, and used the o ptimal parameter combina2 tion of the model as input to t he reg ression support v ecto r machine ( SVR) model. T aking Danjiangko u Reserv oir, the w ater so ur ce of the Middle Route of the South2to2No rth Water Div ersion Project , as the study area, we carried o ut model t ests using the data from 1981 to 2016 in Danjiang ko u. T he results show ed that t he qualificat ion r ate during the model v erificat ion period was 831 33% and the consistency rat e also r eached 831 33% , indicating hig h accuracy and stability. The model has certain r efer2 ence v alue for the annua l runoff fo recast of Danjiang ko u Reser vo ir.

国家重点研发计划( 2016YFC0402201) ;国家自然科学基金( 51709271) ; 青年人才托举工程( 2017QNRC001) ; 中国水利水电科学 研究院基本科研业务费专项( WR0145B212017); 江西省研究生创新专项资金项目( YC2017068)