基于灰色关联分析的模糊支持向量机方法在径流预报中的应用研究

Study and application of fuzzy support vector machine based on gray correlation analysis to streamflow forecasting

针对径流时间序列固有的非线性和随机性特点,提出基于灰色关联分析的模糊支持向量机预报方法。该方法在传统支持向量机任意逼近的非线性映射能力上,引入模糊隶属函数来考虑气候和流域下垫面条件变化下不同时期径流样本对预报结果的影响。预报因子选取是中长期径流预报的一大难点,考虑到相关系数法只能衡量因子间线性相关程度的不足,本文采用灰色关联分析来量化预报因子与预报对象的关联程度,并按关联度大小从众多的相关因子中挑选出对径流过程影响显著的预报因子。将该方法应用于金沙江上游控制站石鼓站的月径流预报中,与GRNN神经网络模型和A-FSVM模型的预报结果比较表明,该方法能提高径流中长期预报的精度,是一种有效的径流时间序列预测模型。

This paper presents a fuzzy support vector machine forecasting method based on gray correlation analysis for forecasting streamflow featured with nonlinearity and randomness. This method takes advantage of traditional SVM in its arbitrary approximation ability and nonlinear mapping, and adopts a fuzzy membership function to consider the impacts of changes in climate and watershed surface conditions on streamflow forecasting results. Predictor selection is difficult in long-term streamflow forecasting and the correlation coefficient method can only measure linear correlation between factors. Hence, we adopt gray correlation analysis to quantify the degree of association and pick out predictors that have significant impact on the streamflow. This model was applied to forecasting of monthly stream flow at Shigu, a control station of the upper Jinsha river. Comparison with the GRNN model and A-FSVM model shows that the method is effective and improves the accuracy of long-term streamflow forecasting.