ARIMA与ANN组合预测模型在中长期径流预报中的应用  

Application of ARIMA-ANN model in the Prediction of Medium and Long-term Runoff

作　　者：傅新忠[1,2] 冯利华[2] 陈闻晨[2] 

机构地区：[1]浙江师范大学文化创意与传播学院 [2]浙江师范大学旅游与资源管理学院,浙江金华321004

出　　处：《水资源与水工程学报》2009年第5期105-109,共5页Journal of Water Resources and Water Engineering

摘　　要：基于时间序列预测模型及BP神经网络,提出了新的组合预测方法。该方法采用三层结构的BP神经网络来构造组合预测模型,运用时间序列模型预测方法得出的预测结果,采用历史滚动法将前5年的预测结果数据作为BP网络的输入,以当前年份的预测结果为网络期望输入,建立了ARIMA-ANN组合预报模型。利用Matlab7神经网络工具箱对塔里木河上游源流卡群水文站的年径流量进行了预报及验证。结果表明:组合模型的预报结果精度高,容错能力强,是中长期径流预报的有效方法。Based on the time series forecasting model and BP neural network, this paper put forward a new combination forecasting method. A new hybrid approach combined by BP Neural Network and ARIMA model were proposed which were constructed by a three-layer structure. This hybrid approach obtained the absolute errors by ARIMA model firstly, then uses the past 5 years data as input values, and uses the BP neural network to simulate the result by a roiling method, finally established the prediction model by the ARIMA with ANN. By assisting the softs of SPSS13 and MatlabT, this paper applies the combination model to predict the annual run- off of the Kaqun hydrological station. The result indicates that the proposed model has higher forecasting accuracy and more tolerant ability. It is an effective model for runoff prediction of medium to long-term.

关 键 词：时间序列 BP神经网络 中长期径流 ARIMA-ANN 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]