摘要
随着深度学习与时间序列传统模型的融合发展,通过利用大量可用数据在整个时间序列集合中估计全局模型的参数,使得传统的局部建模方法得到了实质性的改进。介绍近年来提出的与深度学习相结合的时间序列预测方法及三种时间序列预测模型:深度状态空间模型(DSSM),深度自回归模型(DeepAR),Transformer模型。采用GluonTS时间序列预测框架对上海市出口额数据进行预测并给出效果评估。实验结果表明,基于深度学习的时间序列预测效果明显优于传统的ARIMA模型的预测。
With the development of the integration of deep learning and traditional time series models,the traditional local modeling methods have been substantially improved by using a large amount of available data to estimate the parameters of the global model in the entire time series set.This paper introduces the time series forecasting methods combined with deep learning proposed in recent years and sort out related literature,as well as three kind of time series forecasting models:the deep state space model(DSSM),the deep autoregressive model(DeepAR),and the Transformer model.We used the GluonTS time series forecasting framework to predict the export data of Shanghai.The experimental results show that the time series forecasting effect based on deep learning is significantly better than that of the traditional ARIMA model.
作者
李文
邓升
段妍
杜守国
Li Wen;Deng Sheng;Duan Yan;Du Shouguo(School of Statistics and Information,Shanghai University of International Business and Economics,Shanghai 201620,China;Shanghai Municipal Human Resources and Social Security Bureau Information Technology Center,Shanghai 200051,China)
出处
《计算机应用与软件》
北大核心
2020年第10期64-70,84,共8页
Computer Applications and Software
基金
海关总署决策咨询研究课题(HG-YB009)
国家社会科学基金项目(17BTJ025)
上海市领军人才。
