多种算法寻优LSSVM耦合模型在中长期径流预报中的应用

Application of Multiple Metaheuristic Algorithm-LSSVM Coupling Model in Medium and Long Term Runoff Forecast

针对最小二乘支持向量机(LSSVM)在径流预报问题中相关学习参数难以确定的缺陷,利用近期国内外提出的多种智能优化算法如阿基米德算法(AOA)、美洲雕搜索算法(BES)、黑猩猩优化算法(ChOA)、莱维飞行分布算法(LFD)和鼠群算法(RSO)在有限时间内完成LSSVM模型参数寻优工作,并配合小波包分解技术(WPD)平稳化水文序列,提出多种耦合模型。以雅马渡站和兰西站的年径流量预报为例测试各耦合模型,结果显示:相比较而言,WPD-BES-LSSVM模型综合预报性能最好;BES算法寻优LSSVM模型参数性能更稳健;文中提出的相关模型可为有关中长期径流预报工作提供依据。

In view of the difficulty in determining the relevant learning parameters of least squares support vector machine(LSSVM)in runoff forecasting,several intelligent optimization algorithms such as archimedes optimization algorithm(AOA),bald eagle search algorithm(BES),chimp optimization algorithm(ChOA),levy flight distribution(LFD),and rat swarm optimization algorithm(RSO),are used to optimize the parameters of LSSVM model in limited time.And a variety of coupling models are proposed combined with wavelet packet decomposition(WPD)to stabilize hydrological series.Taking the annual runoff forecast in Yamadu station and Lanxi station as examples to test the models,the results show that WPD-BES-LSSVM model shows the best comprehensive forecast performance.And BES is robust in choosing the parameters of LSSVM model.The models proposed in the paper are able to provide the basis for medium and long term runoff forecast.