基于GRNN模型的鲜猪肉市场日度价格预测与实证研究  

FORECASTING THE MARKET DAILY PRICE OF THE FRESH PORK BASED ON A GENERAL REGRESSION NEURAL NETWORK AND POSITIVE RESEACH

作　　者：楼文高[1,2] 陈芳[2,3,2] 张博[2,3,2] 刘林静[2,3,2] 范晓[1] LOU Wengao CHEN Fang ZHANG Bo LIU Linjing FAN Xiao(Faculty of Finance and Accounting, Shanghai Business School, Shanghai 200235 School of Optical-Electrical and Computer Engineering, The University of Shanghai for Science and Technology, Shanghai 200093 School of Optical-Electrical and Computer Engineering, The University of Shanghai for Science and Technology, Shanghai 200093 School of Optical-Electrical and Computer Engineering, The University of Shanghai for Science and Technology, Shanghai 200093 School of Optical-Electrical and Computer Engineering, The University of Shanghai for Science and Technology, Shanghai 200093 Faculty of Finance and Accounting, Shanghai Business School, Shanghai 200235)

机构地区：[1]上海商学院财经学院,上海200235 [2]上海理工大学光电信息与计算机工程学院,上海200093 [3]上海理工大学出版印刷与艺术设计学院,上海200093

出　　处：《系统科学与数学》2016年第11期1986-1996,共11页Journal of Systems Science and Mathematical Sciences

基　　金：上海高校知识服务平台"上海商贸服务业知识服务中心"建设子项目"上海市价格监管和智能决策系统设计及开发"项目(ZF1226);上海市重点学科"商务经济学"项目资助课题

摘　　要：应用广义回归神经网络(GRNN)模型对上海市某区菜市场2011.3.1-2014.3.25期间鲜猪肉的日度价格(合计732组数据)进行建模预测研究,用逐步减小光滑因子值的办法确定其合理值范围.建模结果表明:训练样本、检验样本和测试样本(简称三类样本)的均方根误差和平均绝对误差非常接近,模型具有较强的泛化能力,绝大部分三类样本的误差都在土0.33元范围内,最大相对误差都小于3%,平均百分比相对误差小于0.45%,预测未来10日鲜猪肉价格的最大绝对误差为0.14元,最大相对误差为0.82%,平均百分比相对误差为0.44%,表明建立的GRNN模型具有很好的鲁棒性、可靠性和较高的预测精度,可用于上海市某区菜市场鲜猪肉日度价格的实际预测,为政府和有关物价部门进行市场调控提供决策依据.The general regression neural network （GRNN） model is applied to fore- cast the daily price of the fresh pork for a given district in Shanghai during 2011.3.1- 2014.3.25 （total 732 samples）. The smooth value-the only one parameter should be determined reasonably-is determined by decreasing the value stepwise. The case study shows that the root-mean-square error （RMSE） and the absolute average error （AAE） of the training data set, the verification data set and the testing data set （three-type data set） are similarly equal. The GRNN＇s performance is good generalized. The errors of the three-type data set are in the range of t0.33 yuan, and the maximum relative errors are less than 3%. The mean absolute percentage errors （MAPEs） of the three-type data set are all less than 0.45%. The absolute error of the forecasted daily price of the fresh pork in the next ten days is 0.14 yuan, and the maximum relative error and MAPE is 0.82% and 0.44%, respectively. These results show that the GRNN model possesses good robustness, reliability and high predicted accuracy, and it can be applied to forecast the daily price of the fresh pork, adjust and control the market by the local government and its price control authorities.

关 键 词：鲜猪肉 日度价格 广义回归神经网络 预测 实证研究 

分 类 号：F224[经济管理—国民经济] F323.7