【摘要】 鉴于降水量数据的高维非线性性和周期性,建立了支持向量回归(SVR)预测模型用于降水量预测,由于对该模型输入特征的选取极为重要,因此提出了一种基于季节自回归(SARI)的输入特征选取方法.利用已有的降水量数据建立SARI模型,通过观察模型表达式提取建立SVR模型所需的输入特征用于训练支持向量机,并通过网格参数寻优法确定SVR模型的参数,进行降水量预测.实例分析中,应用此模型对黄土丘陵半干旱区域的降水量进行预测,将预测结果与季节时间序列(SARIMA)模型的预测结果进行对比,结果表明,模型具有更高的预测精度和拟合优度,可以用于降水量的预测.更多还原

【Abstract】 The annual precipitation datas are time series with high-dimension,nonlinear and periodicity.In view of these characteristics,this study employs the support vector regression(SVR)model to predict precipitation.Since there is no uniform standards in selecting the input features of SVR model when forecasting with precipitation datas as well as few papers have relevant contents,this paper presents a methodology based on seasonal autoregressive(SARI)to obtain the input features of SVR model.First,establish the SARI model based on the existing datas.Then,by observing the expression of SARI model,the Input features for SVR model could be obtained.Finally,according to the input features,some datas are used to train the support vector machine,and the parameters optimization are determined by grid search.Through these steps,the precipitation can be forecasted.An application case study in Semiarid Loess Hilly Region shows that compared with SARIMA model,this model has the better forecasting performance,and it can be used in annual precipitation forecasting.更多还原