摘要
文章将总体经验模态分解(EEMD)方法、长短期记忆(LSTM)模型和Adaboost算法相结合,构建了一个多尺度组合预测模型(EMD-LSTM-Adaboost)。在模型构建过程中,首先采用EEMD方法将商品价格序列分解为不同尺度的本征模态分量(IMF)和一个趋势项。在此基础上,提出采用LSTM神经网络和Adaboost算法相结合的方法对分解后的商品价格序列进行建模和预测,然后集成得到商品价格的预测值。并以沪金为例进行实证分析,结果表明与已有的预测方法相比,文章所提出的EEMD-LSTM-Adaboost方法预测能力更好。
This paper builds a new multi-scale integrated forecasting model by combining Ensemble Empirical Mode Decomposition(EEMD) algorithm with Long Short-Term Memory(LSTM) model and Adaboost algorithm. In the process of modeling,the paper firstly uses EEMD algorithm to decompose the commodity price series into Intrinsic Model Functions(IMF) of different scales and a trend item, on which basis, the method of combining LSTM neural network and Adaboost algorithm is proposed to forecast and model the decomposed commodity price series before obtaining the forecast value of the commodity price by integrating. The paper also takes the futures of Shanghai gold as an example to make an empirical analysis. The results show that the proposed EEMD-LSTM-Adaboost method has better predictive ability than the existing method.
作者
邸浩
赵学军
张自力
Di Hao;Zhao Xuejun;Zhang Zili(Guanghua School of Management, Beijing University, Beijing 100871, China;Harvest Fund Management Co,. Ltd, Beijing 100005, China)
出处
《统计与决策》
CSSCI
北大核心
2018年第13期72-76,共5页
Statistics & Decision
基金
中国博士后科学基金资助项目(2017M611513)