基于PCA-BP神经网络的PM2.5季节性预测方法研究

doi:10.3969/j.issn.1000-2006.201806011

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） ›› 2020, Vol. 44 ›› Issue (5): 231-238.doi: 10.3969/j.issn.1000-2006.201806011

基于PCA-BP神经网络的PM_2.5季节性预测方法研究

张怡文¹(), 郭傲东², 吴海龙¹, 袁宏武¹, 董云春¹

1.安徽新华学院信息工程学院, 安徽合肥 230088
2.重庆邮电大学计算机科学与技术学院,重庆 400065

收稿日期:2018-06-30 修回日期:2020-05-14 出版日期:2020-10-30 发布日期:2020-10-30
基金资助:
安徽省高校优秀青年人才支持计划项目(gxyqZD2018087);安徽省高校自然科学重点项目(KJ2019A0877)

Seasonal prediction of PM_2.5 based on the PCA-BP neural network

ZHANG Yiwen¹(), GUO Aodong², WU Hailong¹, YUAN Hongwu¹, DONG Yunchun¹

1. College of Information Engineering, Anhui Xinhua University, Hefei 230088,China
2. College of Computer Science and Technology, Chongqing University of Posts and Telecommunications, Chongqing 400065,China

Received:2018-06-30 Revised:2020-05-14 Online:2020-10-30 Published:2020-10-30

摘要/Abstract

摘要：

【目的】分季节预测PM_2.5浓度值,利用PCA方法对数据进行降维,分析季节及气象因素对PM_2.5的影响,在提高预测准确率的同时降低时间复杂度。【方法】以合肥市2014—2017年的PM₁₀、SO₂、CO₂、CO、O₃浓度值,以及同时段的气象因素值,对PM_2.5浓度进行预测。数据分析中发现PM_2.5在不同季节浓度差异较大,故本研究选择分季节进行预测;为了提高预测准确率,加入如风力、温度、湿度、气压等气象因素进行预测,同时采用主成分分析(PCA)的方法进行数据降维,将降维后的数据再输入BP神经网络模型进行预测。【结果】实验采用3组实验进行对比:5种污染物指标(PM_2.5-5)预测PM_2.5、加入气象因素的综合12项指标(PM_2.5-12)预测PM_2.5、对综合指标进行PCA处理后的(PM_2.5-PCA)预测PM_2.5。实验结果表明:4个季节的PM_2.5浓度值有较大变化,均方根误差(RMSE)的差值较大;采用PM_2.5-PCA的方法,在任何季节的RMSE均有降低,相关系数(r)均有所提高。【结论】PM_2.5浓度具有季节性特征,采用季节性预测方法可以提高预测准确率;同时采用PCA方法进行降维,可以在保证准确率的同时降低预测时间复杂度。

关键词: PM_2.5, 神经网络, 预测, 主成分分析

Abstract:

【Objective】 We predict the concentration of PM_2.5 during different seasons and use the Principal Component Analysis (PCA) method to reduce the dimensionality of the data, while also improving the accuracy of the prediction and reducing the time complexity, to serve as a reference for travel by people and government decision-making. 【Method】 The PM_2.5 concentration was forecast based on the values of PM₁₀, SO₂, CO₂, CO and O₃ concentration in Hefei from 2014 to 2017, and the meteorological factors during the same period. The data analysis found that the concentration of PM_2.5 varies greatly across seasons; therefore, this study is focused on the forecasting during different seasons. To improve the accuracy of forecasting, influencing meteorological factors such as wind, temperature, humidity and air pressure were added to the forecasting. The PCA method was used for data dimensionality reduction, and then, the data were input into the BP neural network model for prediction. 【Result】 The experiment used three groups of assessments for comparison: five types of pollutant indicators to predict PM_2.5 (PM_2.5-5), addition of twelve comprehensive indicators of meteorological factors to predict PM_2.5 (PM_2.5-12), and use of comprehensive indicators which were processed by PCA (PM_2.5-PCA) to predict PM_2.5. The experimental results showed that the PM_2.5 concentration in the four seasons had large changes, and the difference in Root Mean Square Error(RMSE) is large, when using the PM_2.5-PCA method. We found that if the RMSE is reduced in any season, the correlation coefficient (r) value is increased. 【Conclusion】 The PM_2.5 concentration value has seasonal characteristics, and the seasonal prediction method can improve prediction accuracy. Moreover, adopting the PCA method to reduce data dimensionality can ensure accuracy and decrease the time complexity at the same time.

Key words: PM_2.5, neural networks, prediction, principal component analysis(PCA)

中图分类号:

X831

张怡文,郭傲东,吴海龙,等. 基于PCA-BP神经网络的PM_2.5季节性预测方法研究[J]. 南京林业大学学报（自然科学版）, 2020, 44(5): 231-238.

ZHANG Yiwen, GUO Aodong, WU Hailong, YUAN Hongwu, DONG Yunchun. Seasonal prediction of PM_2.5 based on the PCA-BP neural network[J].Journal of Nanjing Forestry University (Natural Science Edition), 2020, 44(5): 231-238.DOI: 10.3969/j.issn.1000-2006.201806011.

图/表 6

图1

表1

表2

主成分因子载荷矩阵"

季节 season		指标 metrics		Y1	Y2	Y3	Y4	Y5	Y6	Y7
春季 spring		PM₁₀		-0.108 9	0.172 8	-0.331 2	0.684 1	-0.051 1	-0.134 2
		SO₂		0.101 3	0.504 4	-0.054 9	-0.567 5	-0.033 6	-0.007 1
		NO₂		0.168 9	0.270 5	0.472 9	0.108 0	0.421 5	0.019 0
		CO		-0.067 9	-0.585 0	-0.248 5	-0.225 5	0.358 9	0.186 1
		O₃		-0.116 1	0.137 4	-0.634 3	-0.257 4	0.073 6	0.123 7
		平均气压		0.620 1	-0.185 7	-0.297 7	0.145 7	-0.066 1	-0.130 8
		平均气温		0.677 2	-0.167 1	0.106 0	-0.087 5	-0.193 9	0.086 2
		平均相对湿度		0.274 6	0.467 5	-0.301 1	0.113 9	0.165 0	0.191 1
		降水量		0.000 2	-0.001 0	-0.032 8	-0.131 8	0.111 0	-0.585 8
		平均风速		0.099 5	-0.006 4	-0.007 3	0.113 3	0.733 2	0.177 9
		最大风速风向		0.053 2	-0.021 0	-0.088 7	-0.095 2	0.255 4	-0.704 1
夏季 summer		PM₁₀		0.695 3	-0.042 8	-0.160 2	0.267 8	-0.096 4	0.120 1	0.074 5
		SO₂		-0.202 4	0.132 3	0.523 4	-0.327 7	-0.509 9	-0.187 2	-0.018 7
		NO₂		0.040 4	0.606 2	-0.268 0	-0.369 4	0.037 3	-0.079 8	-0.146 4
		CO		-0.601 7	-0.180 0	-0.498 2	0.046 5	0.168 7	0.063 7	0.071 5
		O₃		-0.063 8	-0.627 5	0.018 7	-0.049 4	-0.236 9	-0.045 4	-0.390 7
		平均气压		-0.117 4	0.154 9	0.045 3	0.606 3	-0.361 8	0.066 6	-0.317 0
		平均气温		-0.293 4	0.344 0	0.273 9	0.496 9	0.138 9	0.136 1	0.056 1
		平均相对湿度		0.033 2	0.015 3	-0.077 5	0.191 6	0.148 9	-0.781 8	-0.223 5
		降水量		-0.054 9	-0.040 8	0.036 2	-0.094 3	-0.138 7	0.456 3	0.090 8
		平均风速		-0.016 0	0.199 3	-0.506 7	-0.040 8	-0.509 2	0.093 0	-0.299 4
		最大风速风向		-0.059 6	-0.055 0	-0.197 1	0.134 2	-0.440 4	-0.294 8	0.750 6
季节 season	指标 metrics		Y1		Y2	Y3	Y4	Y5	Y6	Y7
秋季 autumn	PM₁₀		0.330 4		-0.388 5	-0.563 5	0.105 9	0.166 7	-0.224 3	0.007 6
	SO₂		0.408 7		0.306 1	0.501 5	-0.229 5	0.062 5	-0.177 9	-0.152 3
	NO₂		0.187 1		0.514 9	-0.151 0	0.294 2	-0.288 9	0.506 8	0.046 7
	CO		-0.558 9		-0.006 4	0.266 6	0.490 3	0.242 7	-0.149 5	-0.165 4
	O₃		0.225 8		-0.364 2	0.521 1	0.260 0	-0.041 3	0.007 1	0.023 4
	平均气压		0.012 8		0.041 0	0.069 7	0.070 6	0.565 9	0.166 7	0.730 7
	平均气温		-0.017 0		0.545 2	-0.118 9	-0.156 1	0.363 8	-0.404 5	0.006 8
	平均相对湿度		0.562 1		-0.045 1	0.083 2	0.291 1	0.266 4	0.017 9	-0.134 4
	降水量		-0.008 0		-0.043 4	0.120 9	-0.011 8	-0.253 0	0.141 3	0.469 1
	平均风速		0.071 1		0.228 6	-0.151 0	0.646 7	-0.068 5	-0.174 8	-0.031 1
	最大风速风向		0.061 4		0.049 8	0.046 3	0.095 5	-0.481 7	-0.629 7	0.416 7
冬季 winter	PM₁₀		-0.433 3		0.462 5	-0.432 8	0.006 3	-0.309 9	0.019 3	-0.070 1
	SO₂		-0.233 6		-0.428 0	0.426 3	0.163 5	0.068 2	0.263 7	-0.287 8
	NO₂		-0.291 4		-0.033 2	0.413 2	-0.456 2	0.055 5	-0.014 1	0.161 5
	CO		0.659 5		-0.182 5	0.026 7	-0.138 7	-0.467 9	-0.079 9	-0.041 5
	O₃		0.001 0		-0.079 3	0.083 1	0.392 0	-0.485 8	0.149 7	-0.088 8
	平均气压		-0.115 2		-0.518 0	-0.441 4	-0.150 4	-0.242 9	-0.047 1	0.103 5
	平均气温		-0.084 8		-0.479 0	-0.463 2	-0.088 9	0.343 2	0.080 7	0.014 1
	平均相对湿度		-0.439 8		-0.226 1	0.118 4	0.225 0	-0.382 1	-0.097 6	-0.017 7
	降水量		-0.044 0		-0.025 1	-0.011 8	-0.125 2	0.056 0	-0.631 3	-0.747 2
	平均风速		-0.131 0		-0.021 0	0.115 6	-0.653 9	-0.338 6	-0.007 4	0.098 8
	最大风速风向		-0.069 2		-0.110 5	0.125 1	0.260 5	0.027 9	-0.695 9	0.545 3

表2

图2

图3

图4

参考文献 22

[1]	李令军, 王占山, 张大伟 , 等. 2013—2014年北京大气重污染特征研究[J]. 中国环境科学, 2016,36(1):27-35.
	LI L J, WANG Z S, ZHANG D W , et al. Analysis of heavy air pollution episodes in Beijing during 20132014[J]. China Environ Sci, 2016,36(1):27-35.DOI: 10.3969/j.issn.1000-6923.2016.01.005.
[2]	MAKKONEN U, HELLÉN H, ANTTILA P , et al. Size distribution and chemical composition of airborne particles in south-eastern Finland during different seasons and wildfire episodes in 2006[J]. Sci Total Environ, 2010,408(3):644-651. DOI: 10.1016/j.scitotenv.2009.10.050. doi: 10.1016/j.scitotenv.2009.10.050 pmid: 19903567
[3]	杨孝文, 周颖, 程水源 , 等. 北京冬季一次重污染过程的污染特征及成因分析[J]. 中国环境科学, 2016,36(3):679-686.
	YANG X W, ZHOU Y, CHENG S Y , et al. Characteristics and formation mechanism of a heavy winter air pollution event in Beijing[J]. China Environ Sci, 2016,36(3):679-686.DOI: 10.3969/j.issn.1000-6923.2016.03.007.
[4]	张晓茹, 孔少飞, 银燕 , 等. 亚青会期间南京大气PM2.5中重金属来源及风险[J]. 中国环境科学, 2016,36(1):1-11.
	ZHANG X R, KONG S F, YIN Y , et al. Sources and risk assessment of heavy metals in ambient PM2.5 during Youth Asian Game period in Nanjing[J]. China Environ Sci, 2016,36(1):1-11.DOI: 10.3969/j.issn.1000-6923.2016.01.001.
[5]	刘小生, 李胜, 赵相博 . 基于基因表达式编程的PM_2.5浓度预测模型研究[J]. 江西理工大学学报, 2013,34(5):1-5.
	LIU X S, LI S, ZHAO X B . A study on the prediction model of PM_2.5 concentration based on gene expression programming[J]. J Jiangxi Univ Sci Technol, 2013,34(5):1-5.DOI: 10.13265/j.cnki.jxlgdxxb.2013.05.019.
[6]	DUNEAD, POHOATA A, IORDACHE S . Using wavelet-feedforward neural networks to improve air pollution forecasting in urban environments[J]. Environ Monit Assess, 2015,187(7):1-16. DOI: 10.1007/s10661-015-4697-x.
[7]	彭斯俊, 沈加超, 朱雪 . 基于ARIMA模型的PM_2.5预测[J]. 安全与环境工程, 2014,21(6):125-128.
	PENG S J, SHEN J C, ZHU X . Forecast of PM_2.5 based on the ARIMA model[J]. Saf Environ Eng, 2014,21(6):125-128.DOI: 10.13578/j.cnki.issn.1671-1556.2014.06.023.
[8]	贺祥, 林振山 . 基于GAM模型分析影响因素交互作用对PM_2.5浓度变化的影响[J]. 环境科学, 2017,38(1):22-32.
	HE X, LIN Z S . Interactive effects of the influencing factors on the changes of PM_2.5 concentration based on GAM model[J]. Environ Sci, 2017,38(1):22-32.DOI: 10.13227/j.hjkx.201606061.
[9]	刘杰, 杨鹏, 吕文生 , 等. 基于气象因素的PM_2.5质量浓度预测模型[J]. 山东大学学报(工学版), 2015,45(6):76-83.
	LIU J, YANG P, LV W S , et al. Prediction models of PM_2.5 mass concentration based on meteorological factors[J]. J Shandong Univ (Eng Sci), 2015,45(6):76-83.DOI: 10.6040/j.issn.1672-3961.0.2014.214.
[10]	COBOURN W G . An enhanced PM_2.5 air quality forecast model based on nonlinear regression and back-trajectory concentrations[J]. Atmos Environ, 2010,44(25):3015-3023.DOI: 10.1016/j.atmosenv.2010.05.009.
[11]	沈剑波, 雷相东, 李玉堂 , 等. 基于BP神经网络的长白落叶松人工林林分平均高预测[J]. 南京林业大学学报(自然科学版), 2018,42(2):147-154.
	SHEN J B, LEI X D, LI Y T , et al. Prediction mean height for Larix olgensis plantation based on Bayesian-regularization BP neural network[J]. J Nanjing For Univ (Nat Sci Ed), 2018,42(2):147-154.DOI: 10.3969/j.issn.1000-2006.201706012.
[12]	王嫣然, 张学霞, 赵静瑶 , 等. 北京地区不同季节PM_2.5和PM₁₀浓度对地面气象因素的响应[J]. 中国环境监测, 2017,33(2):34-41.
	WANG Y R, ZHANG X X, ZHAO J Y , et al. Study on the response of PM_2.5 and PM₁₀ concentrations to the ground meteorological conditions in different seasons in Beijing[J]. Environ Monit China, 2017,33(2):34-41. DOI: 10.19316/j.issn.1002-6002.2017.02.06.
[13]	姚达文, 刘永红, 丁卉 , 等. 气象参数对基于BP神经网络的PM_2.5日均值预报模型的影响[J]. 安全与环境学报, 2015,15(6):324-328.
	YAO D W, LIU Y H, DING H , et al. Effect of meteorological parameters on the PM_2.5 daily concentration forecasting model based on the BP neural network[J]. J Saf Environ, 2015,15(6):324-328.DOI: 10.13637/j.issn.1009-6094.2015.06.067.
[14]	杨笑笑, 汤莉莉, 张运江 , 等. 南京夏季市区VOCs特征及O₃生成潜势的相关性分析[J]. 环境科学, 2016,37(2):443-451.
	YANG X X, TANG L L, ZHANG Y J , et al. Correlation analysis between characteristics of VOCs and ozone formation potential in summer in Nanjing urban district[J]. Environ Sci, 2016,37(2):443-451.DOI: 10.13227/j.hjkx.2016.02.006.
[15]	陈刚, 刘佳媛, 皇甫延琦 , 等. 合肥城区PM₁₀及PM_2.5季节污染特征及来源解析[J]. 中国环境科学, 2016,36(7):1938-1946.
	CHEN G, LIU J Y, HUANGPU Y Q , et al. Seasonal variations and source apportionment of ambient PM₁₀ and PM_2.5 at urban area of Hefei,China[J]. China Environ Sci, 2016,36(7):1938-1946.DOI: 10.3969/j.issn.1000-6923.2016.07.003
[16]	ZHOU F N, PARK J H, LIU Y J . Differential feature based hierarchical PCA fault detection method for dynamic fault[J]. Neurocomputing, 2016,202:27-35.DOI: 10.1016/j.neucom.2016.03.007.
[17]	REN T, LIU S, MU H P , et al. Temperature prediction of the molten salt collector tube using BP neural network[J]. IET Renew Power Gener, 2016,10(2):212-220.DOI: 10.1049/iet-rpg.2015.0065.
[18]	肖小兵, 刘宏立, 马子骥 . 基于奇异谱分析的经验模态分解去噪方法[J]. 计算机工程与科学, 2017,39(5):919-924.
	XIAO X B, LIU H L, MA Z J . An empirical mode decomposition de-noising method based on singular spectrum analysis[J]. Comput Eng Sci, 2017,39(5):919-924.DOI: 10.3969/j.issn.1007-130X.2017.05.015.
[19]	XIEX, LAM K M . Gabor-based kernel PCA with doubly nonlinear mapping for face recognition with a single face image[J]. IEEE Trans Image Process, 2006,15(9):2481-2492. DOI: 10.1109/TIP.2006.877435. doi: 10.1109/tip.2006.877435 pmid: 16948295
[20]	关蓓蓓, 郑思俊, 崔心红 . 城市人工林空气负离子变化特征及其主要影响因子[J]. 南京林业大学学报(自然科学版), 2016,40(1):73-79.
	GUAN B B, ZHENG S J, CUI X H . The variation and main influencing factors of negative air ions in urban plantation[J]. J Nanjing For Univ (Nat Sci Ed), 2016,40(1):73-79.DOI: 10.3969/j.issn.1000-2006.2016.01.012.
[21]	污染物数据采集[EB /OL].http://www.pm25.in/.2013-08-06/2018-2-20.
[22]	气数据采集[EB /OL].http://www.weather.com.cn/.2008-05-16/2018-5-10.

季节 season	序号 number	特征值 characteristic value	贡献率 contribution	累计贡献率 accumulative contribution	季节 season	序号 number	特征值 characteristic value	贡献率 contribution	累计贡献率 accumulative contribution
春季 spring	1	2.381 9	0.216 5	0.216 5	夏季 summer	1	2.464 2	0.220 9	0.220 9
	2	2.217 8	0.201 6	0.418 2		2	1.897 2	0.170 1	0.391 0
	3	1.549 6	0.140 9	0.559 0		3	1.466 8	0.131 5	0.522 5
	4	1.284 6	0.116 8	0.675 8		4	1.110 4	0.099 5	0.622 0
	5	1.014 2	0.092 2	0.768 0		5	1.044 2	0.093 6	0.715 7
	6	0.961 6	0.087 4	0.855 4		6	0.865 7	0.077 6	0.793 3
	7	0.591 9	0.053 8	0.909 2		7	0.722 7	0.064 8	0.858 1
	8	0.392 8	0.035 7	0.944 9		8	0.608 5	0.054 5	0.912 6
	9	0.323 3	0.029 4	0.974 3		9	0.422 7	0.037 9	0.950 5
	10	0.182 3	0.016 6	0.990 9		10	0.347 4	0.031 1	0.981 7
	11	0.100 0	0.009 1	1.000 0		11	0.204 7	0.018 3	1.000 0
秋季 autumn	1	2.778 2	0.249 2	0.249 2	冬季 winter	1	2.772 8	0.253 4	0.253 4
	2	1.892 3	0.169 7	0.418 9		2	1.740 3	0.159 0	0.412 4
	3	1.241 1	0.111 3	0.530 2		3	1.477 0	0.135 0	0.547 4
	4	1.138 4	0.102 1	0.632 3		4	1.218 1	0.111 3	0.658 7
	5	1.054 7	0.094 6	0.726 9		5	1.010 7	0.092 4	0.751 0
	6	0.936 6	0.084 0	0.810 9		6	0.903 4	0.082 6	0.833 6
	7	0.870 9	0.078 1	0.889 1		7	0.633 0	0.057 8	0.891 4
	8	0.515 1	0.046 2	0.935 3		8	0.533 7	0.048 8	0.940 2
	9	0.311 3	0.027 9	0.963 2		9	0.305 3	0.027 9	0.968 1
	10	0.234 7	0.021 1	0.984 2		10	0.201 2	0.018 4	0.986 5
	11	0.175 8	0.015 8	1.000 0		11	0.147 8	0.253 4	1.000 0