基于粒子群算法优化支持向量机的延河流域水沙模拟

Simulation of Runoff and Sediment Yield in the Yanhe River Basin Based on Support Vector Machine and Particle Swarm Optimization

对标准粒子群算法进行了简化,并基于简化的算法给出了混沌粒子群算法的优化支持向量机算法.基于延河流域甘谷驿水文站1954—1992年的实测年径流和输沙数据、1992—1995实测月径流和输沙数据,利用该算法和常用的几种粒子群支持向量机算法、误差后向传播神经网络算法预测了1993—1997年期间的年径流量和输沙量、1996年的月径流和输沙量.几种算法的预测结果和实测数据进行了比较,通过相对误差、平均相对误差、均方根误差、一致性指标和有效系数等参数,比较了不同算法的预测效果.结果表明,支持向量机算法模拟效果优于神经网络算法,本文提出的基于改进粒子群算法的支持向量机算法的预测效果更好,可用于流域的径流和输沙量的模拟和预报.

The standard support vector machine( SVM) model was simplified and improved by particle swarm optimization. Taking the gauged yearly runoff and sediment yield data from 1954 to 1997 and monthly data from 1993 to 1997 which were gauged at Ganguyi hydrological station in the Yanhe River Basin in the Loess Plateau,the proposed algorithms,together with the widely used SVM algorithms and back propaganda artificial neuro network,were used to simulate and estimate the yearly runoff and sediment yield from 1993 to 1997 and monthly distribution in1996. The estimated results by the different algorithms were compared with actual data in terms of relative error,mean relative error,root-mean-square error,consistency index and coefficient of efficiency. The results demonstrated that SVM algorithms can produce better simulations than artificial neuro network model and the proposed algorithms in this paper can produce the best simulation performance among all the algorithms,therefore,they can be potentially used in runoff and sediment yield forecasting in the basin.