年径流预测的G-SPA模型及其应用

G-SPA Model of Annual Runoff Prediction and Application

提出了以几何形式表示集对的G-SPA模型,该模型以向量之间的夹角、相关系数、欧式距离以及向量的模为指标对径流集合建立对应的指标集合,将指标集合的相似度作为判断径流集合相似度的依据,其优点是不仅考虑了径流大小的相似性,而且考虑了径流变化趋势的相似性。将G-SPA预测模型应用于沱江三皇庙站年均流量预测中,并分别与GRNN神经网络模型以及AR(2)模型的预测结果进行了对比。结果表明:G-SPA模型预测的平均相对误差为16.42%,预测结果优于GRNN模型和AR(2)模型。

One annual runoff prediction model of set pair analysis called G-SPA was proposed in this paper and the'set pair'showed in the form of geometry. Index sets,including the angles、the correlation coefficient、the euclidean distance between vectors and the length of the vectors,were set up in the model,which were corresponding with runoff sets. Then based on the similarity between the index sets,the similarity between the runoff sets could be known. Especially,it took both the similarities of the quantity and the changing trends into account. The G-SPA model was applied to predict the data of annual average runoff series of Sanhuangmiao Station in Tuojiang River,and compared with GRNN model and AR( 2) model. The results show that the predicated mean relative error 16. 42% of G-SPA model is superior to both GRNN model and AR( 2) model.