顾翠伶 王宁 梁艳艳
摘 要:本文对1990-2014年我国人口时间序列进行分析,建立GM(1,1)模型,对未来人口数进行分析,为相关政策的制定提供依据。
关键词:GM(1,1)模型;预测;残差检验;后验差检验;关联度检验
人口预测在政治、经济、环境、教育、医疗卫生、农业生产等方面都有非常重要的应用。人口时间序列预测是根据一个历史的序列观测值,找出符合人口变化规律的函数,根据这个函数将历史观测值作为输入值,预测出未来的人口值。本文对1990-2014年我国人口时间序列进行分析,建立GM(1,1)模型,对未来人口数进行分析,为相关政策的制定提供依据。
1 GM(1,1)模型原理
灰色预测法是一种对不确定性因素的系统进行预测的方法[1],就是对在一定范围内变化的、与时间有关的灰色过程进行预测。灰色时间序列预测是灰色预测的一种,灰色系统常用的数据处理有两种方式,累加和累减两种。
累加是将原始序列通过累加得到生成列。记原始时间序列为:
则一次累加生成列为:
同理可做m次累加,有:
累减是累加的逆运算,累减可将累加生成列还原为非生成列,在建模中获得增量信息。一次累减的公式为:
设时间序列X(0)有n个观察值,X(0)={X(0)(1),X(0)(2),…,X(0)(n)}通过累加生成新序列X(1)={X(1)(1),X(1)(2),…,X(1)(n)},则GM(1,1)模型相应的微分方程为:
求解微分方程即可得到预测模型为:
2 GM(1,1)模型的检验
灰色预测检验一般包括残差检验、关联度检验和后验差检验。
2.1 残差检验。按照预测模型计算,并将累减生成,然后计算原始序列与\1-297\96-x2.jpg>的绝对误差序列及相对误差序列。
2.2 关联度检验。根据关联度的计算方法,计算出\1-297\96-x2.jpg>与原始序列\1-297\96-x2.jpg>的关联系数,然后计算出关联度,根据经验,当ρ=0.5时,关联度大于0.6便满意了。
2.3 后验差检验
计算原始序列的标准差:
计算绝对误差序列的标准差:
计算方差比:
计算小误差概率:
表1 GM(1,1)模型精度检验等级参照表
[\&指标名称\&精度等级\&相对误差\&关联度\&方差比\&小误差概率\&1优
2良好
3合格
4不合格\&0.05
0.10
0.20
0.30\&>0.80
>0.70
>0.60
>0.50\&≤0.35
≤0.50
≤0.65
≥0.65\&≥0.95
≥0.80
≥0.70
<0.0\&]
3 GM(1,1)模型在我國人口序列预测中的应用
这里利用1990-2014年河南省GDP时间序列作为已知序列,建立GM(1,1)模型对未来值进行预测。对原始序列进行累加,得到一次累加生成序列X(1)。通过累加生成序列X(1)建立GM(1,1)模型,利用MATLAB软件进行最小二乘求解,可以得到:
因而灰色预测微分方程为:
化简即可得到预测模型为:
计算拟合值
下面对该模型的预测精度进行检验。实践中可以计算得到绝对误差序列为:
Δ(0)={0,0.3344,0.2735,0.2144,0.1619,0.1102,0.0621,0.0246,
0.0014,0.0201,0.0356,0.0358,0.0348,0.0333,0.0167,0.0189,0.0067,0.0624,0.0183,0.2553,0.1424,0.0748,0.0913,0.1041}
相对误差序列为:
A={0,0.0285,0.0231,0.0179,0.0133,0.0090,0.0050,0.0020,0.0010,
0.0016,0.0028,0.0028,0.0027,0.0026,0.0013,0.0106,0.0055,0.0067,0.0076}
相对误差都小于0.05,预测精度很高。
计算关联度:
min{Δi0}={0,0.3344,0.2735,0.2144,0.1619,0.1102,0.0621,0.0246,
0.0014,0.0201,0.0356,0.0358,0.0348,0.0333,0.0167,0.0189,0.0067,
0.0624,0.0183,0.2553,0.1424,0.0748,0.0913,0.1041}=0
max{Δi0}={0,0.3344,0.2735,0.2144,0.1619,0.1102,0.0621,0.0246,
0.0014,0.0201,0.0356,0.0358,0.0348,0.0333,0.0167,0.0189,0.0067,
0.0624,0.0183,0.2553,0.1424,0.0748,0.0913,0.1041}=0.3344
关联系数为:
由关联系数计算相关系数为:
计算原始序列
所有的ei都小于S0,故P=1,C<0.35,
模型
模型经过检验后可以用于预测,預测公式为:
对未来6年的数值进行预测,结果见表2所示。由预测结果知未来几年我国人口将保持持续增长。
表2 利用GM(1,1)模型对序列未来值进行预测
[年份\&2015\&2016\&2017\&2018\&2019\&2020\&预测值\&13.8667\&13.9515\&14.0369\&14.1227\&14.2092\&14.2961\&]
参考文献:
[1]徐国祥.统计预测和决策[M].上海:上海财经大学出版社,2014.