摘要
GDP预测是经济预测中一个非常重要的问题,随着经济的发展,对其预测精度的要求也越来越高。在考虑样本权重的基础上,提出一种微粒群算法与部分最小二乘回归方法相结合的组合预测方法,即采用微粒群方法对样本最优权重进行求解,在所得样本权重系数的基础上,用部分最小二乘回归方法确定组合预测的权重系数。将该方法用于中国GDP预测取得了较好的结果,与其他几种传统方法相比,预测精度有一定程度的提高,说明算法的有效性和可行性。
GDP forecasting is one of the most important issues in the economic forecasting, along with economic development, demands of its forecast accuracy is becoming higher and higher. Combined forecasting method can effectively integrate the advantages of single forecasting methods and improve prediction accuracy. At present, the representativeness of samples will affect the prediction ability of combining forecasting was not considered in all combining forecasting methods. In this paper, we take the representativeness of sample into account and a new combing forecasting method, PSO-PLS is proposed, we use PSO to search for the best sample weights. PLS is introduced to solving the best weighted average coefficients based on the best sample weights. Finally, we proved the effectiveness and feasibility by applying to GDP of our country, compared with traditional forecasting method, the result of PSO-PLS method is more accurate.
出处
《管理科学》
CSSCI
2008年第3期115-120,F0003,共7页
Journal of Management Science
基金
重庆市自然科学基金(CSTC.2006BB2246)
关键词
微粒群算法
部分最小二乘回归
组合预测
样本权重
GDP
particle swarm optimization
partial least squares
combining forecasting
sample weight
GDP