基于PCA睵SO睸VR的丹江口水库年径流预报研究
2018-12-29张岩杨明祥雷晓辉舒坚牛文生余琅
张岩 杨明祥 雷晓辉 舒坚 牛文生 余琅
摘要: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PCAPSOSVR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8333%、距平一致率也达到8333 %,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。
关键词:年径流预报;主成分分析;粒子群优化算法;回归支持向量机;丹江口水库
中图分类号:P333 文献标志码:A 文章编号:
16721683(2018)05003506
Research on annual runoff forecast of Danjiangkou Reservoir based on PCAPSOSVR
ZHANG Yan1,YANG Mingxiang2,LEI Xiaohui2 ,SHU Jian1,NIU Wensheng3,YU Lang2
(
1.School of Software,Nanchang Hangkong University,Nanchang 330063,China;2.China Institute of Water Resources and Hydropower Research,Beijing 100038,China;3.AVIC Xi′an Institute of Aeronautical Computing Technology,Xi′an 710068,China
)
Abstract:
In the annual runoff forecast,the autocorrelation between meteorological factors will have an impact on forecast accuracy.In order to solve this problem,we added PCA and PSO to the SVR model and established a PCAPSOSVR model.We removed the redundant information and noise,extracted the main features of the factors,and used the optimal parameter combination of the model as input to the regression support vector machine (SVR) model.Taking Danjiangkou Reservoir,the water source of the Middle Route of the SouthtoNorth Water Diversion Project,as the study area,we carried out model tests using the data from 1981 to 2016 in Danjiangkou.The results showed that the qualification rate during the model verification period was 8333% and the consistency rate also reached 8333%,indicating high accuracy and stability.The model has certain reference value for the annual runoff forecast of Danjiangkou Reservoir.
Key words:
annual runoff forecast;principal component analysis;particle swarm optimization;regression support vector machine;Danjiangkou Reservoir
隨着我国水资源调度实践的不断推进,长期径流预报的重要性愈加凸显。然而径流的形成受到降水、蒸发、汇流等一系列复杂过程和人类活动等因素的影响,具有较强的随机性和不确定性,给长期预报带来了困难[1]。随着计算技术与方法的不断发展,如人工神经网络、支持向量机等各种智能学习方法逐渐应用到长期径流预报中。支持向量机(Support Vector Machine,SVM)作为一种新兴的机器学习方法,在20世纪90年代中期被Vapnik等人首次提出[2],主要用于解决分类和回归预测问题。文献[3]在支持向量机建模过程中引入了径向基核函数,简化了非线性问题的求解过程。与人工神经网络模型预报结果比较显示,该模型能提高中长期径流预报的精度。……