针对目前大多数模型均为对已监测区域的空气质量进行预测，而少有对未监测区域进行预测的问题，综合考虑气象因素、空间相关性和时间依赖性对空气质量的影响，提出了一种联合训练模型MLP&ST （MLP&spatial-temporal），模拟预测北京市未监测区域未来时刻的空气质量指数（air quality index，AQI）。通过实验结果对比确定最佳历史时间步长P值为29；然后将模型与其他空气质量预测模型进行对比。实验结果表明，MLP&ST模型在3种评价指标下（RMSE，MAE，MAPE）皆优于其他对比模型，验证了所提出模型具有良好的预测效果。
Air quality prediction system is necessary in guarding human and environment health, and the historical air pollution data collected by monitoring stations in a city and advanced computer equipment makes it possible to forecast air quality in a data-driven way. However, most of the existing methods are suitable for forecasting the air quality in monitored areas, and only few are for unmonitored areas. In this paper, a joint training model MLP&ST (MLP&Spatial-Temporal) is proposed which takes the consideration of the comprehensive impact of meteorological factors, spatial correlation and time dependence on air quality, to predict the future air quality index (AQI) of unmonitored areas in Beijing. Compared with several other air quality forecast models, the optimized historical time step P value of 29 is experimentally obtained. Experimental results show that the hybrid model is superior to other models in performance indices (RMSE, MAE and MAPE) and has good ability of predictability.
[1] Cozzi L. World energy outlook special report 2016:energy and air pollution[J]. International Energy Agency, 2016:17-56.
[2] Beelen R, Hoek G, Van Den Brandt P A, et al. Long-term exposure to traffic-related air pollution and lung cancer risk[J]. Epidemiology (Cambridge, Mass), 2008, 19(5):702-710.
[3] Gan W Q, Koehoorn M, Davies H W, et al. Long-term exposure to traffic-related air pollution and the risk of coronary heart disease hospitalization and mortality[J]. Environmental Health Perspectives, 2011, 119(4):501-507.
[4] 梅波,田茂再.基于时空模型北京市PM2.5浓度影响因素研究[J].数理统计与管理, 2018, 37(4):571-586. Mei B, Tian M Z. Analysis of influential factors on PM2.5 in Beijing based on spatio-temporal model[J]. Journal of Applied Statistics and Management, 2018, 37(4):571-586.(in Chinese)
[5] 周曙东,欧阳纬清,葛继红.京津冀PM2.5的主要影响因素及内在关系研究[J].中国人口·资源与环境, 2017, 27(4):102-109. Zhou S D, Ouyang W Q, Ge J H. Study on the main influencing factors and their intrinsic relations of PM2.5 in Beijing-Tianjin-Hebei[J]. China Population, Resources and Environment, 2017, 27(4):102-109.(in Chinese)
[6] Wen C C, Liu S F, Yao X J, et al. A novel spatiotemporal convolutional long short-term neural network for air pollution prediction[J]. Science of the Total Environment, 2019, 654:1091-1099.
[7] Bey I, Jacob D J, Yantosca R M, et al. Global modeling of tropospheric chemistry with assimilated meteorology:model description and evaluation[J]. Journal of Geophysical Research:Atmospheres, 2001, 106(D19):23073-23095.
[8] Grell G A, Peckham S E, Schmitz R, et al. Fully coupled "online" chemistry within the WRF model[J]. Atmospheric Environment, 2005, 39(37):6957-6975.
[9] Di Carlo P, Pitari G, Mancini E, et al. Evolution of surface ozone in central Italy based on observations and statistical model[J]. Journal of Geophysical Research:Atmospheres, 2007:112(D10).
[10] Castellano M, Franco A, Cartelle D, et al. Identification of NOx and ozone episodes and estimation of ozone by statistical analysis[J]. Water, Air, and Soil Pollution, 2009, 198(1/2/3/4):95-110.
[11] Liu B C, Binaykia A, Chang P C, et al. Urban air quality forecasting based on multidimensional collaborative support vector regression (SVR):a case study of Beijing-TianjinShijiazhuang[J]. PLoS One, 2017, 12(7):e0179763.
[12] Saxena A, Shekhawat S. Ambient air quality classification by grey wolf optimizer based support vector machine[J]. Journal of Environmental and Public Health, 2017, 2017:3131083.
[13] Yu R Y, Yang Y, Yang L Y, et al. RAQ-a random forest approach for predicting air quality in urban sensing systems[J]. Sensors, 2016, 16(1):86.
[14] Hu X F, Belle J H, Meng X, et al. Estimating PM2.5 concentrations in the conterminous United States using the random forest approach[J]. Environmental Science&Technology, 2017, 51(12):6936-6944.
[15] Chattopadhyay S, Chattopadhyay G. Modeling and prediction of monthly total ozone concentrations by use of an artificial neural network based on principal component analysis[J]. Pure and Applied Geophysics, 2012, 169(10):1891-1908.
[16] Feng X, Li Q, Zhu Y J, et al. Artificial neural networks forecasting of PM2.5 pollution using air mass trajectory based geographic model and wavelet transformation[J]. Atmospheric Environment, 2015, 107:118-128.
[17] Ong B T, Sugiura K, Zettsu K. Dynamic pre-training of deep recurrent neural networks for predicting environmental monitoring data[C]//2014 IEEE International Conference on Big Data (Big Data), 2014:760-765.
[18] Fan J, Li Q, Hou J, et al. A spatiotemporal prediction framework for air pollution based on deep RNN[J]. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2017, IV-4/W2:15-22.
[19] Li X, Peng L, Yao X J, et al. Long short-term memory neural network for air pollutant concentration predictions:method development and evaluation[J]. Environmental Pollution, 2017, 231:997-1004.
[20] Zhao J C, Deng F, Cai Y Y, et al. Long short-term memory-fully connected (LSTM-FC) neural network for PM2.5 concentration prediction[J]. Chemosphere, 2019, 220:486-492.
[21] Huang C J, Kuo P H. A deep CNN-LSTM model for particulate matter (PM2.5) forecasting in smart cities[J]. Sensors, 2018, 18(7):2220.
[22] Pak U, Kim C, Ryu U, et al. A hybrid model based on convolutional neural networks and long short-term memory for ozone concentration prediction[J]. Air Quality, Atmosphere&Health, 2018, 11(8):883-895.
