本文基于分解-重构-集成的思想,构建了一个多尺度组合预测模型,选取小麦作为粮食的代表,预测其价格走势。首先,运用集合经验模态分解方法(EEMD)分解价格序列,然后,用灰色关联分析方法对分量序列进行重构,重构为高频、中频、低频和趋势项四个部分,并从不规则因素、季节因素、重大事件和世界经济水平等方面对这四个部分波动特点进行解释,针对不同特点的分量选择不同的方法进行预测,最后对各预测结果用支持向量机集成,并与其他预测模型的预测结果进行比较。实证结果表明,本文构建的多尺度组合模型的预测效果优于灰色预测GM(1,1)、BP神经网络、SVM方法、ARIMA模型等单模型方法和ARIMA-SVM组合模型以及基于EMD和EEMD分解的其他多尺度组合模型。
Forecasting of grain price is an important area of grain market research.In this paper a new multi-scale combined forecasting model was built based on the idea of decomposition-reconstruction-integration.It selected wheat as representative of grain and forecasted its price trend.It used ensembleempirical mode decomposition (EEMD) to decompose price series, then reconstructed the component sequences into high frequency, middle frequency, low frequency and trend sequences with grey correlation method, which can be explained from the angle of irregular factors, seasonal factor, major events and long-term trend.It forecasted different sequences by different methods according to their characteristics, such as BP neural network,Support Vector Machine (SVM), ARIMA and so on.Finally, it integrated prediction results with SVM.The empirical results show that comparing with GM (1, 1), BP neural network, SVM and other single models, ARIMA-SVM combined model as well as other multi-scale model based on EMD or EEMD, multi-scale combined model obtains the best forecast result.
[1] Bates J M, Granger C W J.The combination of forecasts[J].Operational Research Quarterly, 1969, 20(4):451-468.
[2] Krogh A, Vedelsby J.Neural network ensembles, cross validation, and active learning[J].Neural Computing & Applications, 1995, 7(10):231-238.
[3] Antoniadis A, Saptinas T.Wavelets methods for continuous time prediction using Hilbert-valued auto-regressive processes[J].Journal of Multivariate Analysis, 2003, 87(1):133-158.
[4] Kazem A, Sharifi E, Hussain F K, et al.Support vector regression with chaos-based firefly algorithm for stock market price forecasting[J].Applied Soft Computing, 2013,13(2):947-958.
[5] Wang Shouyang, Yu L, Lai K K.A novel hybrid AI system framework for crude oil price forecasting[M]//Shi yong,Xu Weixuan,Chen Zhengxin.Data Mining and knowledging manage ment.Berlin Meidelbng:Springer,2005.
[6] Wang Shonyang, Yu L, Lai K K.Crude oil price forecasting with TEI@I methodology[J].International Journal of Systems Science and Complexity, 2005, 18(2):145-166.
[7] 陈华友.广义加权算术平均组合预测方法的最优化理论基础及性质[J].系统工程理论与实践,2003,23(4):37-41.
[8] 陈华友,盛昭瀚,刘春林.基于向量夹角余弦的组合预测模型的性质研究[J].管理科学学报,2006,9(2):1-8.
[9] 刘轶芳,迟国泰,余方平,等.基于GARCH-EWMA的期货价格预测模型[J].哈尔滨工业大学学报,2006,38(9):1572-1575.
[10] 陈兆荣,雷勋平,王亮,等.基于ARIMA-SVM组合模型的我国农产品价格预测研究[J].财经理论研究,2013,(2):103-107.
[11] Bjorn V.Multiresolution methods for financial time series prediction[C]//Computational Intelligence for Financial Engineering, New York,April 9-11,1995.
[12] Edmundo G, Legey F L, Edmundo A. Forecasting oil price trends using wavelets and hidden Markov models[J]. Energy Economics, 2010, 32(6):1507-1519.
[13] 曹霜,何玉成.基于小波分解的SVM-ARIMA农产品价格预测模型[J].统计与决策,2015,(13):92-95.
[14] Huang N E, Shen Zheng, Long S R, et al.The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis[J].Proceedings of the Royal Society A Mathematical Physical & Engineering Sciences, 1998, 454(1971):903-995.
[15] Yu L, Wang Shouyang, Lai K K. Forecasting crude oil price with an EMD-based neural network ensemble learning paradigm[J].Energy Economics, 2008, 30(5):2623-2635.
[16] 王书平,胡爱梅,吴振信.基于多尺度模型的铜价预测研究[J].中国管理科学,2014,22(8):21-28.
[17] Wu Zhaohua, Huang N E.Ensemble empirical mode decomposition:A noise-assisted data analysis method[J].Advances in Adaptive Data Analysis, 2009, 1(1):1-41.
