基于Prophet-随机森林优化模型的空气质量指数规模预测

详细信息    查看全文 | 推荐本文 |

英文篇名：Scale prediction of AQI based on Prophet-random forest optimization model 作者：常恬君 ; 过仲阳 ; 徐丽丽 英文作者：CHANG Tianjun;GUO Zhongyang;XU Lili;Key Laboratory of Geographic Information Science,Ministry of Education,East China Normal University;School of Geographic Sciences,East China Normal University;East China Sea Forecast Center,State Oceanic Administration; 关键词：Prophet模型 ; 随机森林 ; 时间序列预测 ; 优化模型 英文关键词：Prophet model;;random forest;;time series prediction;;optimization model 中文刊名：环境污染与防治 英文刊名：Environmental Pollution & Control 机构：华东师范大学地理信息科学教育部重点实验室;华东师范大学地理科学学院;国家海洋局东海预报中心; 出版日期：2019-07-15 出版单位：环境污染与防治 年：2019 期：07 基金：国家重点研发计划项目(No.2017YFC0210000) 语种：中文; 页：19-22+27 页数：5 CN：33-1084/X ISSN：1001-3865 分类号：X51

摘要

长时间的规模预测有助于从宏观角度分析事物的发展趋势与规律。对上海市2013—2017年逐日空气质量指数(AQI)进行分析,在此基础上建立了Prophet-随机森林(RF)优化模型。Prophet模型将AQI时间序列趋势分解为趋势项、季节项、节假日效应;RF算法用于弥补Prophet模型无法预测随机非线性部分的缺点,对Prophet模型进行优化,将Prophet-RF优化模型用于AQI的规模预测。结果表明:相比于Prophet模型,Prophet-RF优化模型的预测效果更加精确,其中,拟合值的均方根误差和平均绝对误差均减少了0.161,预测值的均方根误差和平均绝对误差分别减少了0.434和0.399。Prophet-RF优化模型解释性强且精度高,对于时间序列的规模预测具有较明显的优势。
        The long-term scale prediction helps to analyze the development trend and law of things from a macroscopic perspective.Prophet-random forest(RF)optimization model was established on the analysis of the Shanghai daily air quality index(AQI)from 2013 to 2017.The Prophet model decomposed the AQI time series trend into growth items,seasonal and holiday effects.The RF model was used to make up for the defect that the Prophet model could not predict the stochastic nonlinear part,and to optimize the Prophet model.Prophet-RF optimization model was applied to forecast the scale of the AQI.The experimental results showed that compared to the single Prophet model,the prediction results of Prophet-RF optimization model were more accurate.Among them,the root mean square error and mean absolute error of fitting values both decreased by 0.161.Root mean square error and mean absolute error of predictive value decreased by 0.434 and 0.399,respectively.Prophet-RF optimization model had higher precision and was explanatory,having obvious advantages to predict the scale of time series.

引文

[1]刘威.基于支持向量机的城市空气质量时间序列预测模型探究[J].电子测试,2013(20):44-46.
    [2]陈亚玲,赵智杰.基于小波变换和传统时间序列模型的臭氧浓度多步预测[J].环境科学学报,2013,33(2):339-345.
    [3]宋宇辰,甄莎.BP神经网络和时间序列模型在包头市空气质量预测中的应用[J].干旱区资源与环境,2013,27(7):65-70.
    [4]郭利进,井海明,南亚翔,等.基于卡尔曼滤波融合算法的空气质量指数预测[J].环境污染与防治,2017,39(4):388-391.
    [5]TAYLOR S J,LETHAM B.Forecasting at scale[J].The American Statistical,2018,72(1):37-45.
    [6]柴竞,彭草.长沙市2016年节假日与非节假日空气质量指数对比的研究[J].大家健康,2017,11(9):33-34.
    [7]HAN J W,KAMBER M,PEI J.数据挖掘:概念与技术[M].范明,孟小峰,译.3版.北京:机械工业出版社,2012.