基于蚁群算法的支持向量机中长期水文预报模型  

Ant colony optimization-based support vector machine for mid-and-long term hydrological forecasting

作　　者：张俊[1,2] 程春田[1] 申建建[1] 张世钦[3] 

机构地区：[1]大连理工大学水电与水信息研究所,大连116024 [2]浙江电力调度通信中心,杭州310007 [3]福建电力调度通信中心,福州350003

出　　处：《水力发电学报》2010年第6期34-40,共7页Journal of Hydroelectric Engineering

基　　金：国家自然科学基金(50679011)

摘　　要：支持向量机算法(SVM)具有可靠的全局最优性和良好的泛化能力,尤其适用于中长期水文预报中有限样本的学习,然而该算法的成功与否很大程度上取决于其参数的选择,而常规经验选取方法往往不能获得满意效果。本文建立基于蚁群算法参数优化的支持向量机模型(ACO-SVM),利用蚁群算法(ACO)基于进化的随机搜索策略对支持向量机参数进行识别。以福建省安砂水库的月径流预报为例,进行建模仿真,将模拟结果与时间序列方法(ARMA)、人工神经网络方法(BP-ANN)所获得的预报结果进行对比分析。结果表明,在拟合精度方面,ACO-SVM模型相比ARMA模型和BP-ANN模型有不同程度的提高,且增幅较大,且具有较好的泛化性能。Support vector machine（SVM） algorithm is of reliable global optimality and good generalization,suitable for mid-and-long term hydrological forecasting that needs a learning of finite samples.However,the results considerably depend on SVM model parameters and the conventional parameter choosing method by experience is unsatisfactory.This paper proposes an ant colony optimization（ACO）-based SVM model（ACO-SVM） that optimizes the parameters by using an ACO random seeking strategy.A case study of the Ansha reservoir in Fujian province demonstrates the application of ACO-SVM,and the result is compared with those of conventional ARMA and BP-ANN.It is shown that ACO-SVM is much more efficient in global optimization,its forecasting accuracy better than the other two models,and its generalization ability comparable with BP-ANN.

关 键 词：支持向量机 蚁群算法 参数优选 水文预报 

分 类 号：TV124[水利工程—水文学及水资源] P388.1[天文地球—地球物理学]