不同改进的ARIMA模型在水文时间序列预测中的应用  

Application of Different Improved ARIMA Models in the Prediction of Hydrological Time Series

作　　者：杜懿[1,2,3] 麻荣永 DU Yi1,2,3, MA Rongyong1,2,3(1. College of Civil Engineering and Architecture, 2. Key Laboratory of Disaster Prevention and Struc Nanning 530004, Guangxi, China; 3. Guangxi Safety, Guangxi University, Guangxi Nannin rural Safety of Ministry of Ed Key Laboratory of Disaster P g 530004, Guangxi, China; ucation, Guangxi University, revention and Engineering Nanniug 530004, Guangxi, Chin)

机构地区：[1]广西大学土木建筑工程学院,广西南宁530004 [2]广西大学工程防灾与结构安全教育部重点实验室,广西南宁530004 [3]广西大学广西防灾减灾与工程安全重点实验室,广西南宁530004

出　　处：《水力发电》2018年第4期12-14,28,共4页Water Power

基　　金：国家自然科学基金项目(51369005);广西防灾减灾与工程安全重点实验室系统性研究项目(2013ZDX04)

摘　　要：为了拓展ARIMA模型在非线性水文时间序列预测中的应用,对其进行了三种不同形式的改进。为了尽可能地保留数据序列的原始信息,分别将ARIMA模型与Daubechies小波分析、经验模态分解法相结合,建立了WAARIMA和EMD-ARIMA预测模型;由于ARIMA模型对非线性项预测较差,而BP神经网络具有高度的非线性映射能力,将二者进行有效结合,建立了一种BP-ANN-ARIMA预测模型。在南宁市年降水量的预测中表明,改进后的模型较原始模型预测能力有了不同程度的提高,各模型预测的相对误差分别为11.2%、10.1%、6.8%、5.1%。In order to extend the application of ARIMA model in the prediction of nonlinear hydrological time series, three different forms of improvement are proposed here. For retaining the raw information of data sequence as much as possible, the ARIMA model are combined with Daubechies wavelet and EMD method to establish EMD-ARIMA model and WA-ARIMA model respectively. Because of the poor nonlinear prediction of ARIMA model, it is also combined with BP neural network to establish a BP-ANN-ARIMA model. In annual precipitation prediction of Nanning, all improved models have higher prediction accuracy than original model. The relative errors of each model are 11.2%, 10. 1%, 6.8% and 5.1% respectively.

关 键 词：年降水量 ARIMA模型 经验模态分解 小波分析 神经网络 

分 类 号：TV121.7[水利工程—水文学及水资源]