基于参数优化的EEMD-LSSVM年径流组合预测模型

径流预测是水资源管理的基础，其准确性直接影响水资源优化调度的成果。本文针对径流时间序列的内在周期特性，引入一种基于总体经验模态分解（EEMD）的LSSVM组合预测模型，并提出一种基于动态逼近局部搜索粒子群的LSSVM参数寻优方法。基于分解一重构原则，论文首先利用总体经验模式分解法对径流系列进行周期分量提取，然后应用基于参数寻优的LSSVM模型对各分量进行预测和重构。以澧水流域江垭站的年径流预测为例进行模型检验，通过三种预测模型的结果对比，验证了本文组合预测模型的可靠性。

EEMD-LSSVM model of combined forecast of annual runoff based on parameter optimization

Runoff forecast is the base of water resources management and its accuracy has a critical influ- ence on the optimization scheduling of water resources. Aimed at the inner periodic feature of runoff time series, a hybrid annual runoff forecast method based on ensemble empirical mode decomposition （ EE- MD） and least squares support vector machines （LSSVM） was proposed in this paper. To solve the prob- lem of the uncertain parameters of LSSVM, a particle swarm optimization algorithm based on dynamic - approximation research was proposed to optimize the parameter in LSSVM. Based on the principle of de- composition and ensemble, the runoff series is decomposed into lots of periodic components first. And then, the LSSVM model is used to forecast and reconstruct these components of intrinsic mode function. In comparison to the results of three methods in the annual runoff forecast of Jiangya station in Lishui Riv- er, witch shows that the proposed runoff forecast model is reliable.