Research On Ozone Nowcasting Technology In The Yangtze River Delta Region Based On Deep Learning

With the increase of ground level ozone concentration and frequent occurrence of photochemical pollution incidents,ozone has gradually replaced particulate matter as the primary air pollutant in many places.The Yangtze River Delta is economically developed,densely populated,and there are many industrial areas in the region.Therefore,it is very important to study the characteristics of ozone pollution in the Yangtze River Delta and build accurate and effective ozone concentration forecasting models.This paper studies the temporal and spatial distribution characteristics of ground level ozone pollution in the Yangtze River Delta,and establishes a real-time ozone forecast model in the Yangtze River Delta.This article first uses the near-ground ozone observation data from environmental monitoring sites in the Yangtze River Delta from 2017 to 2018 to analyze the change characteristics and spatial distribution characteristics of ground level ozone(O₃)at different time scales,as well as the excesses.The results show that:(1)Compared with 2017,the ground level ozone O₃ concentration in the Yangtze River Delta region increased in 2018,especially in inland cities,and the pollution further increased;Ozone pollution is more serious in summer,followed by spring and the lowest in winter.The daily change of O₃ concentration presents an obvious"single peak structure",which usually reaches the highest value of the day around15:00 in the afternoon.The O₃concentration shows a weekly change in working days higher than weekends,which is the so-called"weekend effect".(2)Compared with 2017,most cities in the region had more days of O₃ concentration exceeding the standard,and O₃ pollution further increased.The O₃ concentration exceeding standard period in the Yangtze River Delta is mainly concentrated at 13:00-18:00,especially at 14:00-16:00.Secondly,the impact of meteorological factors,precursors and particulate matter on the ground level ozone concentration is analyzed using the observation data(weather data and environmental data)of the Yangtze River Delta region.The results show that the ground level ozone concentration is positively correlated with air temperature and is negatively correlated with relative humidity.The correlation between precipitation and ozone concentration is not obvious.The relationship between wind speed and ozone concentration is more complicated.It is not a simple linear relationship.NO₂ concentration and CO concentration are negatively correlated with ozone concentration,and the relationship between the concentration of particulate matter(PM_2.5 and PM₁₀)and ozone concentration is also more complicated,which needs to be further explored.Finally,based on the Encoder-Forecaster architecture,this paper proposes a sequence-to-sequence deep learning model that introduces the Attention mechanism to predict ground level ozone,and through retrospective prediction,real-time forecast and compared with WRF-Chem to verify the prediction performance.The results show that the Seq2seq model can reliably predict the ozone concentration with high accuracy,single-step prediction can be used for accurate value prediction,and multi-step prediction can be used for prediction of the evolution trend of ozone concentration.The root mean square error of the 1-hour ozone forecast on the test data set of the model is 12.40?g/m³,and the average absolute error of the 1-hour ozone forecast is 9.27?g/m³.