大数据 ›› 2019, Vol. 5 ›› Issue (6): 101-110.doi: 10.11959/j.issn.2096-0271.2019053

• 应用 • 上一篇    

基于SARIMA-LSTM的门诊量预测研究

卢鹏飞1,须成杰2,张敬谊1,韩侣3,李静1   

  1. 1 万达信息股份有限公司,上海 201112
    2 复旦大学附属妇产科医院,上海 200090
    3 长春理工大学,吉林 长春 130022
  • 出版日期:2019-11-15 发布日期:2020-01-10
  • 作者简介:卢鹏飞(1991- ),男,万达信息股份有限公司大数据产品部数据挖掘工程师,主要研究方向为医疗数据挖掘、机器学习、光谱分析|须成杰(1983- ),男,复旦大学附属妇产科医院信息科工程师,主要研究方向为大数据及人工智能、互联网医疗+物联网、医院信息无纸化管理、医疗可信云计算等|张敬谊(1974- ),女,博士,万达信息股份有限公司大数据产品部总经理、教授级高级工程师,主要研究方向为并行计算、智能分析、城市信息化等|韩侣(1992- ),男,长春理工大学硕士生,主要研究方向为统计机器学习、数据挖掘|李静(1973- ),女,万达信息股份有限公司大数据产品部资深产品经理、高级工程师,主要研究方向为医疗卫生大数据、健康医疗大数据+人工智能
  • 基金资助:
    上海市科委民生科技支撑计划专项临床医学科技创新项目(17411950500);上海市科委民生科技支撑计划专项临床医学科技创新项目(17411950505)

Research on the prediction of outpatient volume based on SARIMA-LSTM

Pengfei LU1,Chengjie XU2,Jingyi ZHANG1,Lyu HAN3,Jing LI1   

  1. 1 Wonders Information Co.,Ltd.,Shanghai 201112,China
    2 Gynecology Hospital of Fudan University,Shanghai 200090,China
    3 Changchun University of Science and Technology,Changchun 130022,China
  • Online:2019-11-15 Published:2020-01-10
  • Supported by:
    Special Clinical Medical Science and Technology Innovation Project of Livelihood Science and Technology Support Plan of Shanghai Science and Technology Commission(17411950500);Special Clinical Medical Science and Technology Innovation Project of Livelihood Science and Technology Support Plan of Shanghai Science and Technology Commission(17411950505)

摘要:

为了实现更加稳健和精准的门诊量预测,构建了一种基于SARIMA-LSTM的门诊量预测模型。该方法首先使用SARIMA模型对门诊量进行单指标建模,提取门诊量指标蕴含的周期、趋势等信息,然后构建了以节日天数、法定上班天数、平均最高气温等多个相关指标为输入的多对一LSTM模型,对SARIMA模型残差进行进一步学习,实现残差与多个变量间的非线性关系抽取。实证结果表明,构建SARIMA-LSTM混合模型相较5种主流预测方法具有更高的一步预测精度,具有较好的实际应用价值。

关键词: 季节性差分自回归滑动平均模型, 长短期记忆网络, 门诊预测, 残差

Abstract:

In order to achieve more robust and accurate outpatient volume prediction,a hybrid prediction model based on SARIMALSTM was constructed.SARIMA model was used to build a single index model of outpatient volume to extract the cycle,trend and other information contained in outpatient volume index.Then multiple related indexes,including holiday days,legal working days,average maximum temperature,were used as input of a many-to-one LSTM model,in order to further learn the residual of SARIMA model and extract the nonlinear relationship between residual and multiple variables.The empirical results show that the SARIMA-LSTM hybrid model constructed in this paper has higher prediction accuracy than the five mainstream prediction methods,so it has good practical application value.

Key words: seasonal auto-regressive integrated moving average model, long short term memory, outpatient forecast, residual

中图分类号: 

No Suggested Reading articles found!