基于EMD和CatBoost算法的改进时间序列模型——以大连市PM2.5预测为例

doi:10.12302/j.issn.1000-2006.202205005

国家林草科技领军期刊
中国精品科技期刊
中国高校百佳科技期刊
江苏省新闻出版政府奖期刊奖
RCCSE林学权威期刊（A+）
CSCD核心期刊
Scopus数据库收录期刊
中文核心期刊
SCD核心期刊

作者加群：102861116

微信公众号：南京林业大学学报

高级检索

南京林业大学学报（自然科学版） ›› 2024, Vol. 48 ›› Issue (3): 268-274.doi: 10.12302/j.issn.1000-2006.202205005

基于EMD和CatBoost算法的改进时间序列模型——以大连市PM_2.5预测为例

赵凌霄¹^,²(), 李智扬³, 屈磊磊⁴^,^*()

1.大连海洋大学海洋与土木工程学院,辽宁大连 116023
2.复旦大学大气与海洋科学系,上海 200438
3.重庆大学土木工程学院,重庆 400044
4.大连海洋大学信息工程学院,辽宁大连 116023

收稿日期:2022-05-04 修回日期:2023-08-11 出版日期:2024-05-30 发布日期:2024-06-14
通讯作者: *屈磊磊(quleilei@dlou.edu.cn),副教授。
作者简介:赵凌霄(LingxiaoZ@outlook.com)。
基金资助:
辽宁省博士科研启动基金项目(2020-BS-216);国家级大学生创新创业训练计划(202110158002);辽宁省大学生创新创业训练计划(S202210158006)

Improved time series models based on EMD and CatBoost algorithms: taking PM_2.5 prediction of Dalian City as an example

ZHAO Lingxiao¹^,²(), LI Zhiyang³, QU Leilei⁴^,^*()

1. College of Marine and Civil Engineering, Dalian Ocean University, Dalian 116023, China
2. Department of Atmospheric and Oceanic Sciences, Fudan University, Shanghai 200438, China
3. College of Civil Engineering, Chongqing University, Chongqing 400044, China
4. College of Information Engineering, Dalian Ocean University, Dalian 116023, China

Received:2022-05-04 Revised:2023-08-11 Online:2024-05-30 Published:2024-06-14

摘要/Abstract

摘要：

【目的】解决传统大气PM_2.5浓度时序预测时精度较低问题,减少PM_2.5时间序列的非线性、高噪声、不平稳与波动性对预测的影响,从而更精确地预测PM_2.5浓度。【方法】以2014年1月1日至2022年1月31日大连市雾霾天气时PM_2.5数据为例,提出了经验模态分解(EMD)、分类提升 (CatBoost)、自回归综合移动平均模型(ARIMA)组合的混合机器学习时间序列模型,并与传统自回归模型(AR)、ARIMA,以及只加入EMD方法后的混合模型进行比较。【结果】混合模型EMD-CatBoost-ARIMA较原始序列均方根误差(RMSE)改进20.76%,平均绝对值误差(MAE)改进17.40%,希尔不等系数(TIC)改进29.17%。【结论】对于高熵值的重构序列,EMD分解方法和CatBoost算法能够显著提升PM_2.5时间序列模型的预测性能。相比较传统时间序列模型,EMD-CatBoost-ARIMA模型对大气PM_2.5浓度预测性能较高。

关键词: PM_2.5浓度, 经验模态分解(EMD), 时间序列模型, 混合模型, CatBoost算法, 机器学习, 大连市

Abstract:

【Objective】 The study aims to address the problem of low accuracy in traditional PM_2.5 concentration time series prediction, and to reduce the impact of nonlinearity, high noise, instability and volatility on the prediction of PM_2.5 time series, to predict PM_2.5 concentration more accurately. 【Method】 The haze PM_2.5 data of Dalian City from January 1, 2014 to January 31, 2022 was used as an example. In this study, a hybrid machine learning time series model with the combination of empirical modal decomposition (EMD), classification boosting (CatBoost) and autoregressive integrated moving average model (ARIMA) was proposed. It was compared with the traditional autoregressive model (AR), ARIMA and the hybrid model with only the EMD method. 【Result】 The hybrid model EMD-CatBoost-ARIMA improved the root mean square error (RMSE) of the original sequence by 20.76%, the mean absolute error (MAE) by 17.40%, and the theil inequality coefficient (TIC) by 29.17%. 【Conclusion】 For reconstructed sequences with high entropy values, the EMD decomposition method and CatBoost algorithm can significantly improve the prediction performance of PM_2.5 time series models. Compared with the traditional time series models, the EMD-CatBoost-ARIMA model has higher performance in PM_2.5 concentration prediction.

Key words: PM_2.5 concentration, empirical modal decomposition(EDM), time series model, hybrid model, CatBoost algorithm, machine learning, Dalian City

中图分类号:

X513

赵凌霄,李智扬,屈磊磊. 基于EMD和CatBoost算法的改进时间序列模型——以大连市PM_2.5预测为例[J]. 南京林业大学学报（自然科学版）, 2024, 48(3): 268-274.

ZHAO Lingxiao, LI Zhiyang, QU Leilei. Improved time series models based on EMD and CatBoost algorithms: taking PM_2.5 prediction of Dalian City as an example[J].Journal of Nanjing Forestry University (Natural Science Edition), 2024, 48(3): 268-274.DOI: 10.12302/j.issn.1000-2006.202205005.

图/表 8

图1

表1

图2

图3

图4

表2

图5

图6

参考文献 23

[1]	World health organization. Ambient air pollution:a global assessment of exposure and burden of disease[M]. Geneva: World Health Organization, 2016.
[2]	TAO R J, CAO W J, LI M H, et al. PM_2.5 compromises antiviral immunity in influenza infection by inhibiting activation of NLRP3 inflammasome and expression of interferon-Β[J]. Mol Immunol, 2020, 125:178-186.DOI: 10.1016/j.molimm.2020.07.001.
[3]	DE MARCO A, AMOATEY P, KHANIABADI Y O, et al. Mortality and morbidity for cardiopulmonary diseases attributed to PM_2.5 exposure in the metropolis of Rome,Italy[J]. Eur J Intern Med, 2018, 57:49-57.DOI: 10.1016/j.ejim.2018.07.027.
[4]	GUO L C, LV Z L, MA W J, et al. Contribution of heavy metals in PM_2.5 to cardiovascular disease mortality risk,a case study in Guangzhou,China[J]. Chemosphere, 2022, 297:134102.DOI: 10.1016/j.chemosphere.2022.134102.
[5]	施婷婷, 王帅, 杨立娟, 等. 中国华东地区PM_2.5浓度时空变化及与景观格局关联研究[J]. 遥感技术与应用, 2024, 39(2): 435-446.
	SHI T T, WANG S, YANG L J, et al. The spatial-temporal change of PM_2.5 concentration and its relationship with landscape pattern in East China[J]. Remote Sensing Technology and Application, 2024, 39(2): 435-446.DOI:10.11873/j.issn.1004-0323.2024.2.0435.
[6]	汪伟舵, 吴涛涛, 张子振. 基于ARIMA模型的杭州市PM_2.5预测[J]. 哈尔滨师范大学自然科学学报, 2018, 34(3):49-55.
	WANG W D, WU T T, ZHANG Z Z. Forecast of PM_2.5 in Hangzhou based on ARIMA model[J]. Nat Sci J Harbin Norm Univ, 2018, 34(3):49-55.DOI: 10.3969/j.issn.1000-5617.2018.03.009.
[7]	CHYON F A, SUMAN M N H, FAHIM M R I, et al. Time series analysis and predicting COVID-19 affected patients by ARIMA model using machine learning[J]. J Virol Methods, 2022, 301:114433.DOI: 10.1016/j.jviromet.2021.114433.
[8]	杨茜雯, 朱萌. 基于ARIMA模型对扬州市PM_2.5的分析和预测[J]. 黑龙江环境通报, 2022, 35(1):35-37,40.
	YANG Q W, ZHU M. Analysis and prediction of PM_2.5 in Yangzhou based on ARIMA model[J]. Heilongjiang Environ J, 2022, 35(1):35-37,40.DOI: 10.3969/j.issn.1674-263X.2022.01.012.
[9]	彭斯俊, 沈加超, 朱雪. 基于ARIMA模型的PM2.5预测[J]. 安全与环境工程, 2014, 21(6): 125-128.
	PENG S J, SHEN J C, ZHU X, et al. Forecast of PM_2.5 based on the ARIMA model[J]. Safety and Environ Engine, 2014, 21(6): 125-128. DOI: 10.13578/j.cnki.issn.1671-1556.2014.06.023.
[10]	严宙宁, 牟敬锋, 赵星, 等. 基于ARIMA模型的深圳市大气PM_2.5浓度时间序列预测分析[J]. 现代预防医学, 2018, 45(2):220-223,242.
	YAN Z N, MOU J F, ZHAO X, et al. The time series prediction of PM_2.5 in Shenzhen based on ARIMA model[J]. Mod Prev Med, 2018, 45(2):220-223,242.
[11]	谢心庆, 郑薇, 开璇, 等. 基于时间序列和多元方法的乌鲁木齐PM_2.5浓度分析[J]. 云南大学学报(自然科学版), 2016, 38(4):595-601.
	XIE X Q, ZHENG W, KAI X, et al. An analysis of PM_2.5 concentration based on time sequence and multivariate methods in Urumqi City[J]. J Yunnan Univ (Nat Sci Ed), 2016, 38(4):595-601.DOI: 10.7540/j.ynu.20150789.
[12]	余辉, 袁晶, 于旭耀, 等. 基于ARMAX的PM_2.5小时浓度跟踪预测模型[J]. 天津大学学报(自然科学与工程技术版), 2017, 50(1):105-111.
	YU H, YUAN J, YU X Y, et al. Tracking prediction model for PM_2.5 hourly concentration based on ARMAX[J]. J Tianjin Univ (Sci Technol), 2017, 50(1):105-111.DOI: 10.11784/tdxbz201504033.
[13]	吴明晖, 张广洁, 金苍宏. 基于多模态信息融合的时间序列预测模型[J]. 计算机应用, 2022, 42(8):2326-2332.
	WU M H, ZHANG G J, JIN C H. Time series prediction model based on multimodal information fusion[J]. J Comput Appl, 2022, 42(8):2326-2332.DOI: 10.11772/j.issn.1001-9081.2021061053.
[14]	何泽森. 移动APP日活跃用户量预测研究[D]. 杭州: 浙江工商大学, 2018.
	HE Z S. Research on forecasting the mobile APP daily active user[D]. Hangzhou: Zhejiang Gongshang University, 2018.
[15]	WONGSATHAN R, CHANKHAM S. Improvement on PM-10 forecast by using hybrid ARIMAX and neural networks model for the summer season in Chiang Mai[J]. Procedia Comput Sci, 2016, 86:277-280.DOI: 10.1016/j.procs.2016.05.062.
[16]	ALADAG E. Forecasting of particulate matter with a hybrid ARIMA model based on wavelet transformation and seasonal adjustment[J]. Urban Clim, 2021, 39:100930.DOI: 10.1016/j.uclim.2021.100930.
[17]	张棋. 基于机器学习的中国气象干旱时空预测研究[D]. 郑州: 华北水利水电大学, 2021.
	ZHANG Q. Study of meteorological drought spatiotemporal forecast methods in China based on machine learning[D]. Zhengzhou: North China University of Water Resources and Electric Power, 2021.
[18]	LU C G, ZHANG S A, XUE D, et al. Improved estimation of coalbed methane content using the revised estimate of depth and CatBoost algorithm:a case study from southern Sichuan basin,China[J]. Comput Geosci, 2022, 158:104973.DOI: 10.1016/j.cageo.2021.104973.
[19]	DING Y, CHEN Z Q, LU W F, et al. A CatBoost approach with wavelet decomposition to improve satellite-derived high-resolution PM_2.5 estimates in Beijing-Tianjin-Hebei[J]. Atmos Environ, 2021, 249:118212.DOI: 10.1016/j.atmosenv.2021.118212.
[20]	OMIDVARNIA A, MESBAH M, PEDERSEN M, et al. Range entropy:a bridge between signal complexity and self-similarity[J]. Entropy (Basel), 2018, 20(12):962.DOI: 10.3390/e20120962.
[21]	王涯鑫, 李捷辉, 王健. 甲醇-柴油双燃料发动机甲醇泄漏故障预诊断研究[J]. 车用发动机, 2022(1):86-92.
	WANG Y X, LI J H, WANG J. Pre-diagnosis of methanol leakage fault for methanol-diesel dual fuel engine[J]. Veh Engine, 2022(1):86-92.DOI: 10.3969/j.issn.1001-2222.2022.01.014.
[22]	ZHENG J M, HU M X, WANG C H, et al. Spatial patterns of residents’ daily activity space and its influencing factors based on the CatBoost model:a case study of Nanjing,China[J]. Front Archit Res, 2022, 11(6):1193-1204.DOI: 10.1016/j.foar.2022.04.003.
[23]	TAYLOR K E. Summarizing multiple aspects of model performance in a single diagram[J]. J Geophys Res, 2001, 106(D7):7183-7192.DOI: 10.1029/2000jd900719.

DF检验参数 Dickey-Fuller test parameter	值 value
统计量test statistic	-5.67
P	8.93×10^-7
滞后项lags used	25.0
观测数据量number of observations used	2.93×10³
1%关键值1%critical value	-3.43
5%关键值5%critical value	-2.86
10%关键值10%critical value	-2.57
数据类别data type	float64

分类 classification	模型 model	均方根误差 RMSE	平均绝对值误差 MAE	希尔不等系数 TIC
原始时间序列模型 original time series model	AR	16.577 6	10.389 4	0.325 9
原始时间序列模型 original time series model	ARIMA	15.733 0	10.970 4	0.272 4
EMD混合模型 EMD hybrid model	EMD-AR	13.895 7	9.458 1	0.248 4
	EMD-ARIMA	13.881 3	9.419 3	0.246 5
	EMD-ARIMA-AR	13.880 9	9.463 4	0.248 0
CatBoost综合模型 CatBoost integrated model	EMD-CatBoost-AR	13.143 2	9.059 8	0.231 1
CatBoost综合模型 CatBoost integrated model	EMD-CatBoost-ARIMA	13.135 3	9.060 8	0.230 8

基于EMD和CatBoost算法的改进时间序列模型——以大连市PM_2.5预测为例

Improved time series models based on EMD and CatBoost algorithms: taking PM_2.5 prediction of Dalian City as an example

RichHTML

PDF