Earth Science |
|
|
|
|
Hourly concentration prediction of PM2.5 based on RNN-CNN ensemble deep learning model |
Jie HUANG1,2, Feng ZHANG1,2, Zhenhong DU1,2, Renyi LIU1,2, Xiaopei CAO1,2 |
1.Zhejiang Provincial Key Lab of GIS, Zhejiang University, Hangzhou 310028, China 2.Department of Geographic Information Science, Zhejiang University, Hangzhou 310027, China |
|
|
Abstract Most of the current PM2.5 prediction models show unstable prediction effect and weak generalization ability. This research aims to design a prediction model called RNN-CNN for PM2.5 hourly concentration prediction based on ensemble deep learning. We choose Recurrent Neural Network (RNN) with strong memory and Convolutional Neural Network (CNN) with strong feature expression ability as individual learners and choose Stacking, an ensemble learning technique, to combine RNN and CNN so that we can take full advantages of both in the forecast. RNN-CNN can not only use the contextual information on the timeline to predict the future concentration ,but also extract different levels of essential features from the high dimensional features for prediction, ensuring the stability of the forecast. We take the air quality data of 1 466 monitoring stations in mainland China in 2016 as samples to compare the performance of RNN-CNN with the individual learners RNN and CNN. Experiment results show that RNN-CNN performs better and achieves higher prediction accuracy and stronger generalization ability than the individual learners RNN and CNN, What’s more, its index of agreement on the 34% of test stations is higher than 0.97, which indicates that the ensemble deep learning model RNN-CNN can be effectively used for prediction of PM2.5 hourly concentration at large scales.
|
Received: 23 January 2018
Published: 25 May 2019
|
|
基于RNN-CNN集成深度学习模型的PM2.5小时浓度预测
针对目前大部分PM2.5预测模型预测效果不稳定、泛化能力不强的现状,以记忆能力较强的循环神经网络(RNN)和特征表达能力较强的卷积神经网络(CNN)为基础,采取Stacking集成策略对两者进行融合,提出了RNN-CNN集成深度学习预测模型。该模型不仅充分利用时间轴上的前后关联信息去预测未来的浓度,而且在不同层次上将自动提取的高维时序数据通用特征用于预测,以保证预测结果的稳定性。最后,对集成之前的RNN、CNN和集成之后的RNN-CNN模型,以2016年中国大陆地区1 466个监测站点的空气质量数据为样本进行实例验证,结果表明,RNN-CNN在PM2.5时间序列预测上的表现明显优于集成之前的RNN和CNN,而且泛化误差更低,在34%站点上的拟合度超过0.97,该模型可用于大范围区域的PM2.5小时浓度预测。
关键词:
PM2.5小时浓度预测,
RNN,
CNN,
深度学习,
集成学习
|
|
1 SORENSENM, DANESHVARB, HANSENM, et al.Personal PM2.5 exposure and markers of oxidative stress in blood[J]. Environmental Health Perspectives, 2003, 111(2): 161-166. DOI: 10.1289/ehp.111-1241344 2 GIBSONM D, KUNDUS, SATISHM.Dispersion model evaluation of PM2.5, NOx, and SO2, from point and major line sources in Nova Scotia, Canada using AERMOD Gaussian plume air dispersion model[J]. Atmospheric Pollution Research, 2013, 4(2):157-167. DOI:10.5094/APR.2013.016 3 WUY, GROSSB M, MOSHARYF. Assessing satellite AOD based and WRF/CMAQ output PM2.5 estimators[C]// SPIE Defense, Security, and Sensing,2013:872319. DOI: 10.1117/12.2027430 4 KLOOGI, NORDIOF, COULLB A, et al.Incorporating local land use regression and satellite aerosol optical depth in a hybrid model of spatio-temporal PM2.5 exposures in the mid-Atlantic states[J]. Environmental Science & Technology, 2012, 46(21):11913. DOI: 10.1021/es302673e 5 XUW, HUANGZ C, ZHANGQ N. Prediction and interpolation of PM2.5 based on space-time model[J]. Journal of Jiangsu Normal University(Natural Science Edition), 2016,34(3):70-75. DOI: 10.3969/j.issn.2095-4298.2016.03.016 6 LIX, ZHANGC, LIW, et al.Evaluating the use of DMSP/OLS nighttime light imagery in predicting PM2.5 concentrations in the Northeastern United States[J]. Remote Sensing, 2017, 9(6): 620.doi:10.3390/rs9060620 7 HEQ, HUANGB.Satellite-based mapping of daily high-resolution ground PM2.5 in China via space-time regression modeling[J]. Remote Sensing of Environment, 2018, 206: 72-83. DOI: 10.1016/j.rse.2017.12.018 8 LYU B, HUY, CHANGH H, et al.Improving the accuracy of daily PM2.5 distributions derived from the fusion of ground-level measurements with aerosol optical depth observations, a case study in North China[J]. Environmental Science & Technology, 2016, 50(9): 4752. DOI: 10.1021/acs.est.5b05940 9 YINQ, WANGJ, HUM, et al.Estimation of daily PM2.5 concentration and its relationship with meteorological conditions in Beijing[J]. Journal of Environmental Sciences, 2016, 48(10): 161-168.doi:10.1016/j.jes.2016.03.024 10 HOUJ X, LIQ, ZHUY J, et al.Real-time forecasting system of PM2.5 concentration based on spark framework and random forest model[J]. Science of Surveying & Mapping, 2017, 42(1):1-6. DOI: 10.16251/j.cnki.1009-2307.2017.01.001 11 ONG B T, SUGIURAK, ZETTSUK.Dynamically pre-trained deep recurrent neural networks using environmental monitoring data for predicting PM2.5[J]. Neural Computing and Applications, 2016, 27(6): 1553-1566. DOI:10.1007/s00521-015-1955-3 12 LIT, SHENH, YUANQ, et al. Estimating ground‐level PM2.5 by fusing satellite and station observations: A Ge‐intelligent deep learning approach[J]. Geophysical Research Letters, 2017, 44: 1-9. DOI: 10.1002/2017GL075710 13 FANJ X, LIQ, ZHUY J, et al. A spatiotemporal prediction framework for air pollution based on deep RNN[J]. Science of Surveying and Mapping,2017,42(7):76-83.DOI: 10.16251/j.cnki.1009-2307.2017.07.013 14 LECUNY, BENGIOY, HINTONG. Deep learning[J]. Nature, 2015, 521(7553): 436-444. DOI: 10.1038/nature14539 15 HOCHREITERS, SCHMIDHUBERJ. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735. DOI: 10.1162/neco.1997.9.8.1735 16 CHO K, MERRIENBOER BVAN, BAHDANAUD, et al.On the properties of neural machine translation: Encoder-decoder approaches[J]. Computer Science, 2014.arxiv:1409.1259. 17 CHENY, JIANGH, LIC, et al.Deep feature extraction and classification of hyperspectral images based on convolutional neural networks[J]. IEEE Transactions on Geoscience & Remote Sensing, 2016, 54(10): 6232-6251. DOI: 10.1109/TGRS.2016.2584107 18 SRIVASTAVAN, HINTONG, KRIZHEVSKYA, et al.Dropout: A simple way to prevent neural networks from overfitting[J]. Journal of Machine Learning Research, 2014, 15(1): 1929-1958. 19 TIELEMANT, HINTONG.RMSProp: Divide the gradient by a running average of its recent magnitude[J]. Neural Networks for Machine Learning, 2012(4): 26-31. 20 LECUNY. Generalization and network design strategies[J]. Connectionism in Perspective, 1989:143-155. 21 BENGIOY I, GOODFELLOWJ, COURVILLEA.Deep Learning[M]. Cambridge: The MIT Press, 2016. 22 SPRINGENBERGJ T, DOSOVITSKIYA, BROXT, et al. Striving for simplicity: The all convolutional net[J]. Eprint Arxir, 2014:1-14. 23 DIETTERICHT G.Ensemble methods in machine learning[J]. Lecture Notes in Computer Science, 2000, 1857(1): 1-15. DOI:10.1007/3-540-45014-9_1 24 WOLPERTD H.Stacked Generalization[M]. Boston: Springer US, 2011.doi:10.1016/s0893-6080(05)80023-1 25 WILLMOTTC J.On the validation of models[J]. Physical Geography, 1981, 2(55): 184-194. |
|
Viewed |
|
|
|
Full text
|
|
|
|
|
Abstract
|
|
|
|
|
Cited |
|
|
|
|
|
Shared |
|
|
|
|
|
Discussed |
|
|
|
|