应用DRNN和ARIMA组合模型的时空集成预测方法  

Integrated Spatio-temporal Forecasting Method by DRNN and ARIMA Combined Model

作　　者：梅志雄[1,2] 

机构地区：[1]华南师范大学地理科学学院,广东广州510631 [2]中山大学地理科学与规划学院,广东广州510275

出　　处：《小型微型计算机系统》2010年第4期657-661,共5页Journal of Chinese Computer Systems

基　　金：国家杰出青年科学基金项目(40525002)资助;广东省自然科学基金项目(5005940)资助

摘　　要：在评析目前时空预测研究现状的基础上,提出基于动态回归神经网络(DRNN)和自回归集成移动平均(ARIMA)组合模型的时空集成预测方法.该方法先用ARIMA模型对时空数据的时序进行预测,再用DRNN捕获时空数据间隐藏的空间关系,最后用线性回归将二者整合起来,得到集成预测结果.案例实验结果表明:该方法比不考虑空间影响的预测方法或单一的预测方法有更高的精度;该方法具有良好的动态处理和计算能力,对跨空间的动态过程的预测有效可行.After reviewing the current spatio-temporal forecasting researches,this paper proposes a spatio-temporal integration forecasting approach based on dynamic recurrent neural network（DRNN） and autoregressive integrated moving average（ARIMA） combined model.The approach first forecasts time series by ARIMA model,then captures the hidden spatial correlation by DRNN between spatio-temporal data,and finally combines the individual temporal and spatial forecasting results based on linear regression to produce the final spatio-temporal integration forecasting result.Experimental results show that the approach can obtain better performance and forecasting precision than the approaches that don＇t consider spatial correlation or monomial forecasting method,can forecast dynamic process over space effectively in virtue of its strongly dynamic handling and computing abilities.

关 键 词：动态回归神经网络 ARIMA模型 集成 时空预测 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]