一种基于渐消因子与滑动窗口的小区域人口预测方法

A Small-area Population Forecasting Method Based on Fading Factor and Sliding window

对小区域人口数据的插补方法、平滑方法与预测方法进行详细研究,并针对小区域人口时间序列数据结构变化快的特点,提出了一种基于渐消因子与滑动窗口的预测方法。该方法不对现有人口预测模型的结构进行任何改变,只将人口年平均增长率、地区人口规模比例、人口年平均增长率之差等模型参数视为动态变量。通过滑动窗口不断引入新的预测值更新模型参数,同时利用渐消因子不断强化新增信息作用,弱化历史信息影响,提高参数的时效性与模型的灵活性。以湖南省澧县32个乡镇1998—2015年人口数据为研究对象,开展了大量的人口预测与分析。结果显示:1.附加人口总量约束信息的人口预测模型总体预测精度要高于未使用人口总量信息的纯数学模型;2.通过数据平滑处理可减少偶然因素影响,预测精度可提高5%~8%;3.采用本文所提方法可显著改善现有模型的人口预测能力,预测未来5年和10年的人口数量,预测精度可分别提高18%和33%。

The demographic model is widely regarded as the 'gold model' for the large-area population forecasting. However,demographic forecasting techniques do not perform particularly well for small areas. In this study,we propose a new approach for small-area population forecasting that borrows strength from the fading factor and sliding time window. The innovation series are continually introduced by the sliding window to update the model parameters. Meanwhile,the fading factor is used to strengthen the role of new information and weaken the impact of historical information. Therefore,it is more suitable for the rapid changing of small-area population than the existing methods. To verify the validation of this method,the population data of 32 villages and towns in Lixian County,Hunan province,China,from 1998 to 2015,is adopted and a large number of experiments are performed. The results show that: 1. the models with the constraint condition of total population are more accurate than those without the constraint information; 2.the data smooth technology can reduce the influence of accidental errors and improve the prediction accuracyby 5%-8%; 3. the proposed method can significantly improve the forecasting ability of the existing models. The prediction accuracy is increased by 18% and 33% respectively when predicting the population size in the next five years and ten years.