The application of deep learning method in Shanghai PM2.5 prediction
MA Jing-hui1,2,3, CAO Yu1,3, YU Zhong-qi1,3, QU Yuan-hao1,3, XU Jian-ming1,3
1. Yangtze River Delta Center for Environmental Meteorology Prediction and Warning, Shanghai 200030, China;
2. Fudan University, Shanghai 200433, China;
3. Shanghai Key Laboratory of Meteorology and Health, Shanghai 200030, China
In order to improve the forecasting ability of PM2.5 concentration, especially the forecasting ability of PM2.5 heavy pollution, this study was based on the mesoscale meteorological-chemical coupled modeling system (WRF-Chem) forecast data in combination with weather and PM2.5 (fine particulate matter) observational data, a machine learning model of PM2.5 prediction in Shanghai, China was established. The results showed that the machine learning algorithm combined with WRF-Chem prediction obviously corrected the deviation of model prediction due to the non-objectivity of the model, and improved the prediction effect. The linear regression method (Lasso) could not obtain a suitable optimization effect, and the deep learning sequence to sequence algorithm was selected to improve the model and verified by experiments. The overall correlation coefficient of the method for the PM2.5 prediction increased from 0.51 to 0.79. The root mean square error was reduced from 25.9μg/m3 to 15.0μg/m3. The Seq2seq modified PM2.5 forecasting model based on WRF-Chem can effectively improve the prediction accuracy. It took a useful application prospect in air quality forecast.
Jordan M I. Serial order:A parallel distributed processing approach[J]. Advances in psychology, 1997,121:471-495.
[2]
胡译文.基于WRF-Chem的气象资料同化对气溶胶模拟的影响[C]//.中国气象学会.第35届中国气象学会年会S12大气成分与天气、气候变化与环境影响暨环境气象预报及影响评估,中国气象学会, 2018:3. Hu Y W. Impact of meteorological data assimilation on aerosol simulation based on WRF-Chem[C]//Chinese Meteorological Society. 35th Annual Meeting of Chinese Meteorological Society S12Atmospheric Composition and Weather, Climate Change and Environmental Impact Environmental Meteorological Forecast and Impact Assessment, Chinese Meteorological Society, 2018:3.
[3]
唐孝炎,张远航,邵敏.大气环境化学[M]. 2版.北京:高等教育出版社, 2006:447-449. Tang X Y, Zhang Y H, Shao M. Atmospheric environmental chemistry-[M]. 2nd Edition. Beijing:Higher Education Press, 2006:447-449.
[4]
王占山,李晓倩,王宗爽,等.空气质量模型CMAQ的国内外研究现状[J]. 环境科学与技术, 2013,(S1):386-391. Wang Z S, Li X Q, Wang Z S, et al. Current status of air quality model CMAQ at home and abroad[J]. Environmental Science and Technology, 2013,(S1):386-391.
[5]
薛文博,王金南,杨金田,等.国内外空气质量模型研究进展[J]. 环境与可持续发展, 2013,38(3):14-20. Xue W B, Wang J N, Yang J T, et al. Research progress on air quality model at home and abroad[J]. Environment and Sustainable Development, 2013,38(3):14-20.
[6]
Yumimoto K, Uno I. Adjoint inverse modeling of CO emissions over Eastern Asia using four-dimensional variational data assimilation[J]. Atmospheric Environment, 2006,40(35):6836-6845.
[7]
李小飞,张明军,王圣杰,等.中国空气污染指数变化特征及影响因素分析[J]. 环境科学, 2012,33(6):1936-1943. Li X F, Zhang M J, Wang S J, et al. Analysis of the characteristics and influencing factors of China's air pollution index[J]. Environmental Science, 2012,33(6):1936-1943.
[8]
朱玉强.几种空气质量预报方法的预报效果对比分析[J]. 气象, 2004,10:30-33. Zhu Y Q. Comparative analysis of forecasting effects of several air quality forecasting methods[J]. Meteorology, 2004,10:30-33.
[9]
Yumimoto K, Uno I. Adjoint inverse modeling of CO emissions over Eastern Asia using four-dimensional variational data assimilation[J]. Atmospheric Environment, 2006,40(35):6836-6845.
[10]
孟凯,程兴宏,徐祥德,等.基于CMAQ源同化反演方法的京津冀局地污染源动态变化特征模拟研究[J]. 环境科学学报, 2017,37(1):52-60. Meng K, Cheng X H, Xu X D, et al. Simulation of dynamic variation characteristics of land-based pollution sources based on CMAQ source assimilation inversion method[J]. Chinese Journal of Environmental Science, 2017,37(1):52-60.
[11]
Xu X, Xie L, Cheng X, et al. Application of an adaptive nudging scheme in air quality forecasting in China[J]. Journal of Applied Meterology and Climatology, 2008,47(8):2105-2114.
[12]
靳璐滨,臧增亮,潘晓滨,等.PM2.5和PM2.5~10资料同化及在南京青奥会期间的应用试验[J]. 中国环境科学, 2016,36(2):331-341. Jin L B, Zang Z L, Pan X B, et al. PM2.5 and PM2.5-10 data assimilation and application test during Nanjing Youth Olympic Games[J]. China Environmental Science, 2016,36(2):331-341.
[13]
任万辉,苏枞枞,赵宏德.城市环境空气污染预报研究进展[J]. 环境保护科学, 2010,3:9-11. Ren W H, Su Z Z, Zhao H D. Research progress on urban air pollution forecasting[J]. Environmental Protection Science, 2010,3:9-11.
[14]
常涛.支持向量机在乌鲁木齐大气污染预报中的应用研究[C]//中国气象学会会, 2008. Chang T. Application of support vector machine in air pollution forecasting in Urumqi[C]//Chinese Meteorological Society, 2008.
[15]
Bedoui S, Gomri S, Samet H, et al. A prediction distribution of atmospheric pollutants using support vector machines, discriminant analysis and mapping tools (case study:tunisia)[J]. Journal of Physical Chemistry C, 2016,114(36):15516-15521.
[16]
Just A C, Carli M M, Shtein A, et al. Correcting measurement error in satellite aerosol optical depth with machine learning for modeling PM2.5 in the Northeastern USA[J]. Remote Sens-Basel, 2018,10.
[17]
Zhong R, Coauthors. Forecasting hand, foot, and mouth disease in Shenzhen based on daily level clinical data and multiple environmental factors[J]. Biosci. Trends, 2018,12:450-455.
[18]
吴喜之.复杂数据统计方法-基于r的应用(第三版)[M]. 北京:中国人民大学出版社, 2015:41-56. Wu X Z. Complex data statistics method-application based on r (Third Edition)[M]. Beijing:Renmin University of China Press, 2015:41-56.
[19]
周广强,谢英,吴剑斌,等.基于WRF-Chem模式的华东区域PM2.5预报及偏差原因[J]. 中国环境科学, 2016,36(8):2251-2259. Zhou G Q, Xie Y, Wu J B, et al. Based on WRF-Chem model, the PM2.5forecast and deviation reason in East China region[J]. Chinese Environmental Science, 2016,36(8):2251-2259.
[20]
Sutskever I, Vinyals O, Le Q V. Sequence to sequence learning with neural networks[Z]. 2014,4:3104-3112.
[21]
Emmons L, Walters S, Hess P,et al. Description and evaluation of the Model for Ozone and Related chemical Tracers, version 4(MOZART-4)[J]. Geosci. Model Dev., 2010,3:43-67.
[22]
Elman J L. Finding structure in time[J]. Cognitive Sci., 1990, 14:179-211.
[23]
Bengio Y, Simard P, Frasconi P, et al. Learning long-term dependencies with gradient descent is difficult[J]. Transactions on Neural Networks, 1994,5:157-166.
[24]
Graves A. Long short-term memory[M]. Supervised Sequence Labelling with Recurrent Neural Networks:Springer Berlin Heidelberg, 2012:1735-1780.
[25]
Cho K, Van M B, Gulcehre C, et al. Learning phrase representations using RNN Encoder-Decoder for statistical machine translation[Z]. Computer Science, In Proceedings of the ACL Conference on Empirical Methods in Natural Language Processing (EMNLP), 2014:1724-1734.
[26]
Tibshirani R. Regression shrinkage and selection via the lasso[J]. Journal of the Royal Statistical Society. Series B. Methodological, 1996,58(1):267-288.