| With the rapid development of social economy and the continuous expansion of land scale,the meteorological conditions,air quality and visibility of a city are constantly changing.Among them,visibility is a meteorological factor that has an important impact on traffic travel,production and military operations.Poor visibility and bad weather pose a great threat to the safety of people's lives and property Therefore,the research on visibility is of great significance for ensuring traffic safety,promoting economic development and promoting pollution prevention and controlIn this paper,the core city Shanghai,which is relatively deficient in visibility research in recent years,is selected as the research area,and the spatial and temporal variation characteristics of visibility in this area in 2016-2017 are discussed using meteorological monitoring data and numerical prediction data,and the influencing factors of visibility are explored.On this basis,the deficiencies of existing visibility prediction research were summarized,and the advantages of other research methods were learned.A visibility prediction model based on the convolution neural network which can extract spatial correlation features and a gated recurrent unit with time memory was constructed,and its prediction performance was tested through different dimensions.The main research results are summarized as follows(1)On the basis of filling in a small number of missing visibility values with K nearest neighbor algorithm,the visibility was statistically analyzed in three time dimensions of interannual,monthly and daily.The results showed that the average visibility in 2017 was improved by 2255m compared with that in 2016.The seasonal change features are significant,the visibility is good in summer and autumn,poor in spring and winter,the maximum and minimum values of average visibility appear in August and January respectively.The average visibility during the day showed a periodic change pattern of rising from 6 a.m.to 15 p.m.and falling from 15 p.m.to 6 a.m.on the next day.By kriging interpolation to simulate the whole area of visibility change trend,present the characteristic of increasing northwest to the southeast.(2)According to the correlation analysis method,the correlation coefficient between visibility and meteorological factors and air pollutants was obtained.The results showed that visibility was negatively correlated with relative humidity,PM2.5 concentration and a variety of air pollutants,and positively correlated with wind speed,temperature and ozone concentration.Among them,relative humidity and PM2.5 concentration are the most important influencing factors of visibility in this region.In spring,summer and autumn,visibility is most affected by relative humidity,and in winter by PM2.5 concentration.In addition,the fitting curve of PM2.5 concentration and visibility under different relative humidity shows that when either relative humidity and PM2.5 concentration is at a high value,the change of the other has little impact on visibility.(3)The meteorological,atmospheric pollutants and high-altitude forecast data of all stations at each time were reconstructed into time series data with spatial grid structure,which were then input into the built CNN-GRU visibility prediction model.The model was optimized by using the momentum improved stochastic gradient descent algorithm,and the prediction effect of this model was compared with other methods.The experimental results show that the prediction effect of the established deep learning fusion prediction model is better than that of other single models and the weather research and forecasting model.The absolute absolute error and root-mean-square error of Shanghai city as a whole are 190.05 m and 239.09 m respectively,reaching a very high accuracy of current visibility prediction.In addition,the experiment of changing the time step also shows that the model can predict the visibility in the next 12 hours with high accuracy. |
PDF Full Text Request