基于灰色预测和神经网络的人口预测  

POPULATION FORECAST BASED ON COMMBINATION OF GRAY FORECAST AND ARTIFICIAL NEURAL NETWORKS

作　　者：赖红松[1] 祝国瑞[2] 董品杰[1] 

机构地区：[1]温州市国土资源局,中国浙江温州325027 [2]武汉大学资源与环境科学学院,中国湖北武汉430079

出　　处：《经济地理》2004年第2期197-201,共5页Economic Geography

摘　　要：人口预测是土地利用总体规划的重要基础工作。未来人口规模是土地利用总体规划中确定各类土地需求量控制性指标,调整土地利用结构,实现土地供需平衡,解决人地矛盾的重要依据。人口预测是否科学准确,直接关系到总体规划方案是否合理和实用。利用灰色预测建模所需信息少、方法简单的特点和神经网络具有较强的非线性映射能力的特性,提出一种基于灰色预测和神经网络的人口预测方法。首先对人口规模的NARMA(p,q)的递归网络模型进行一步预测及其灰色预测GM(1,1)等维新息模型预测,然后再用前馈神经网络对GM(1,1)模型和递归网络模型的预测值进行组合预测以作为其最终的预测值。以温州市为例,对其总人口进行了试验预测。结果表明:NARMA(p,q)递归网络模型比GM(1,1)模型具有更高的预测精度,而FNN模型的组合预测效果优于其它单一预测模型。Population forecast is an important foundation work in General Land Use Planning(GLUP). Future total population is the important basis of deciding the controlling indexes of the needs of lands, adjusting the land use structure for realizing the equilibrium of land supply and demand, and solving the contradiction between population and land in GLUP. Whether population forecast is scientific and accurate is directly relation to whether the GLUP scenario is reasonable and practical. With the advantage of gray forecasting method, which is simple and needs less modeling information, and artificial neural networks, which possesses the characteristics of strong nonlinear fitting, based on the combination of gray forecast and artificial neural networks a new method of population forecasting is put forward in this paper. The single-step forecaster, recurrent neural network for NARMA(p,q) modeling ,and the gray system forecaster, equal dimension and new information model of GM(1,1) , are run firstly in parallel. Then the two individual forecasts of the single-step forecaster and the gray system forecaster are mixed by feed forward neural network (FNN) and the results as the final forecast of the population. The performance is tested on the population forecasting for Wenzhou city, Zhejiang. The results indicate that the single-step forecaster has better accuracy than the gray system forecaster and the FNN combination module outperforms the one of its the single-step forecaster and the gray system forecaster.

关 键 词：人口预测 灰色预测GM(1 1)模型 人口规模 前馈神经网络模型 FNN模型 递归网络模型 RNN模型 

分 类 号：C923[社会学—人口学]