摘要
对1978-2011年重庆市生产总值时间序列进行分析,研究发现四阶差分之后,数据趋于平稳;四阶差分序列的自相关系数一阶截尾,最终建立时间序列ARIMA(1,4,1)模型,并测算残差平方和,但ARIMA模型的残差序列存在自相关,对拟合效果产生了影响;基于时间序列ARIMA(1,4,1)模型研究的基础,进一步采用线性神经网络对序列进行学习和仿真计算,结果表明:神经网络的模拟效果优于ARIMA时间序列的模拟效果.
This paper analyzes the time series of Chongqing GDP during 1978-2011,the research finds that the data after fourth-order difference are approximately stationary, that the first-order of autocorrelation coefficient of fourth-order difference sequence is truncated, as a result, the time series ARIMA ( 1,4, 1 ) Model is set up, the residual square sum is calculated, however, the residual sequence of ARIMA Model has autocorrelation and affects fitting effect.Based on the research on time series ARIMA ( 1,4,1 ) Model, by further using linear neural network to conduct learning and simulation on the sequence, the result indicates that the simulation effect of neural network is better than that of ARIMA time series.
出处
《重庆工商大学学报(自然科学版)》
2014年第2期37-42,共6页
Journal of Chongqing Technology and Business University:Natural Science Edition
基金
国家"十一五"科技支撑计划重大项目(2006BAJ05A06)
重庆市科委重点攻关项目(2008AC0043)
重庆工商大学创新型项目(yjscxx2013-026-09)