支持向量机在半干旱半湿润地区水文预报中的应用研究

Applied Research on Support Vector Machine in Hydrological Forecast in Semi-arid and Semi-humid Area

支持向量机是近年来提出的一种机器学习新算法,采用结构风险最小化准则,把学习问题转化为一个二次规划问题来获得最优解,克服了BP神经网络方法中无法避免的局部极值问题。根据支持向量机的原理和方法,在介绍半干旱半湿润地区汾河水库上游流域自然及水文特性的基础上,建立了基于支持向量机的径流预测模型,利用1990-2005年的水文资料进行了检验,并与BP神经网络预测结果进行了对比,结果表明,支持向量机预报模型的精度比BP神经网络有提高,且月径流和非汛期日径流中的预报结果可以用于指导实践。

Support Vector Machine（SVM） is a new machine-learning algorithm proposed recently.It transfers the learning problem into a second planning to acquire the optimal solution according to the principle of structure risk minimum,and it overcomes the shortcoming of falling into local minimum of neural network.Based on introducing natural and hydrological characteristics of the Fenhe reservoir basin,the runoff forecast model based on the SVM according to its principle and method is constructed.The hydrological data during 1990-2005 are adapted to verify the model,and the results are compared with those of BP neural network.The results show that the runoff forecast precision of SVM is higher than that of BP neural network,and the forecast results of monthly runoff and daily runoff in non-flood season can guide the practice management of the water resources.