基于机器学习的两个代表城市上呼吸道感染与气象要素关系及其发病风险预测研究  

A Study on the Relationship between Upper Respiratory Tract Infection and Meteorological Elements and the Outbreak Risk Prediction in Two Representative Cities Using Machine Learning

作　　者：郑甲炜 王式功 尹立[4] 吴千鹏 张祥健 杨燕[4] 黄开龙 ZHENG Jiawei;WANG Shigong;YIN Li;WU Qianpeng;ZHANG Xiangjian;YANG Yan;HUANG Kaiong(College of Atmospheric Sciences/Institution of Environmental Meteorology and Health,Chengdu University of Information Technology,Chengdu 610225,China;Institute of Arid Meteorology,China Meteorological Administration,Lanzhou 730020,China;CMA-CUIT Joint Laboratory of Meteorology and Environment on Health,Chengdu 610225,China;Meteorological Medicine Center of Panzhihua Central Hospital,Panzhihua 617000,China;Shantou Meteorological Bureau,Shantou 515000,China)

机构地区：[1]成都信息工程大学大气科学学院/环境气象与健康研究院,四川成都610225 [2]中国气象局兰州干旱气象研究所,甘肃兰州730020 [3]中国气象局—成都信息工程大学气象环境与健康联合实验室,四川成都610225 [4]攀枝花市中心医院气象医学研究中心,四川攀枝花617000 [5]汕头市气象局,广东汕头515000

出　　处：《沙漠与绿洲气象》2023年第4期160-168,共9页Desert and Oasis Meteorology

基　　金：海南省南海气象防灾减灾重点实验室开放基金项目(SCSF202007);2021年省级科技计划转移支付专项资金项目(21ZYZF-S-01);攀枝花市科学技术局创新中心建设项目(2021ZX-5-1);2020年度第三批攀枝花市市级科技计划项目(2020CY-S-5);2021年度第二批攀枝花市市级科技计划项目(2021CY-S-4)。

摘　　要：选取华南地区深圳市、西南地区攀枝花市2个不同气候区的当地医院上呼吸道感染发病逐日就诊病例数据和同期气象数据,采用随机森林方法和RNN(Recurrent neural network)深度学习方法,通过对两地上呼吸道感染发病特征及其与气象条件关系进行研究,分别构建了两地上呼吸道感染发病风险预测模型。结果表明:(1)深圳市上呼吸道感染就诊人数峰值出现在6—8月,谷值出现在1—2月,呈现以热不舒适的效应为主;而攀枝花市上呼吸道感染就诊人数峰值出现在11月—次年1月,谷值出现在3—7月,呈现以冷不舒适效应为主。(2)逐日平均气温的变化对两地上呼吸道感染发病的影响最明显,当日平均气温>25℃或者<10℃时,两地上呼吸道感染发病风险明显上升。(3)日平均风速影响次之,它与日平均相对湿度和日平均气温一起,通过对气候舒适度产生影响,进而影响人群上呼吸道感染发病情况。(4)在上呼吸道感染与气象要素关联性分析及预测方法优选的基础上,基于RNN深度学习方法构建的两城市上呼吸道感染发病风险预测模型,可为当地相关疾病风险预测及防控提供重要科技支持。The data of daily patients of upper airway cough syndrome in local hospitals and meteorological data in the same period were selected in Shenzhen and Panzhihua from two different climate zones.By using random forest method and RNN(Recurrent neural network) deep learning method,the risk prediction models of upper airway cough syndrome in the two cities were constructed by studying the characteristics of upper airway cough syndrome and its relationship with meteorological elements.The results showed that:(1)The peak value of patients with upper airway cough syndrome in Shenzhen appeared from June to August every year,and the valley value appeared from January to February,showing the hot discomfort effect.The peak value of patients with upper airway cough syndrome in Panzhihua appeared from November to January of the following year,and the valley value appeared from March to July every year,showing the cold discomfort effect.(2) The daily average temperature had a significant impact on the incidence of upper airway cough syndrome,and the risk increased significantly when the temperature was above 25 ℃ or below 10 ℃.(3) The daily average wind speed,together with daily average relative humidity and the daily average temperature,affected the incidence of upper airway cough syndrome by the climate comfort effect.(4)Based on the analysis of the correlation between upper airway cough syndrome and meteorological elements,as well as the optimization of prediction methods,the risk prediction model of upper airway cough syndrome in two cities based on RNN deep learning method can provide important scientific and technological support for local related disease risk prediction and prevention and control.

关 键 词：上呼吸道感染 气象条件 滞后响应关系 随机森林模型 RNN深度学习模型 

分 类 号：P49[天文地球—大气科学及气象学]