基于ARIMA-Intervention-LSTM的组合预测模型  

Hybrid forecasting model based on ARIMA-Intervention-LSTM

作　　者：冯思曼 闫亮 张艳辉 蔡霞[3] FENG Siman;YAN Liang;ZHANG Yanhui;CAI Xia(Faculty of Science,Beijing University of Technology,Beijing 100124,China;School of Mathematics and Statistics,Hebei University of Economics and Business,Shijiazhuang,Hebei 050061,China;School of Science,Hebei University of Science and Technology,Shijiazhuang,Hebei 050018,China)

机构地区：[1]北京工业大学理学部,北京100124 [2]河北经贸大学数学与统计学学院,河北石家庄050061 [3]河北科技大学理学院,河北石家庄050018

出　　处：《河北工业大学学报》2023年第2期28-34,共7页Journal of Hebei University of Technology

基　　金：国家自然科学基金(12001155);河北省自然科学基金(A2020207006,A2022208001);河北省统计科学研究计划项目(2022HY16);河北省高等教育教学改革研究与实践项目(2022GJJG166)。

摘　　要：针对某时刻存在异常的序列数据,首先建立添加异常值或干预效应的ARIMA (Autoregressive Integrated Moving Average)模型,之后应用LSTM (Long-Short Term Memory)模型对ARIMA模型残差序列进行深度学习。通过对波动较为明显的股票收盘价格日度数据和受“新冠”疫情影响的公路货运量序列数据进行实证分析,证实该模型在对某时刻发生不同程度突变的试验数据进行预测时,能够明显提高预测精度。Since the sequence data with anomalies appear at a certain time,an ARIMA(Autoregressive Integrated Mov⁃ing Average)model adding outliers or intervention effects is established,and then apply LSTM(long short term memory)model to carry out deep learning on the residual sequence of ARIMA model.Through the empirical analysis of the daily data of stock closing price with obvious noise and the series data of highway freight volume affected by the intervention of COVID-19,it is confirmed that the model can significantly improve the prediction accuracy when predicting the test data with different degrees of mutation at a certain time.

关 键 词：ARIMA 干预分析 LSTM 组合模型 预测 

分 类 号：C812[社会学—统计学] F224.7