Study On Temporal And Spatial Variation Characteristics And Prediction Of Expressway Visibility In Wuxi Area

Low visibility weather is one of the most serious meteorological disasters in expressway traffic,which poses a great threat to the national economy and social life.There is little research on the occurrence characteristics and prediction methods of low visibility weather in Wuxi area.Based on the meteorological observation data during the period from January 1^st in 2016 to December 31^st in 2019 collected from the national meteorological stations network and the Traffic Meteorological Automatic Monitoring System in Wuxi area and the reanalyzed meteorological data from NCEP（National Center of Environmental Prediction in USA）at the same time,in this paper,the temporal and spatial change pattern of the visibility in Wuxi area was analyzed and its meteorological influencing factors were discussed.A Bi-directional LSTM visibility prediction model based on bidirectional long-short term memory and 1-D CNN-GRU visibility prediction model based on one-dimensional convolution neural network model and the gated cyclic unit model were constructed.The main results were showed as follows:（1）The occurrence frequency of low visibility weather in Wuxi area and its expressways displayed a downward trend year by year,and the visibility had obvious seasonal,monthly and daily variation characteristics.In terms of seasonal variation,the average visibility was the highest in summer and the lowest in winter.On monthly variation,the average visibility was the highest in August and the lowest in January.The lowest value of visibility in a day appears around 05:00 before sunrise,and the highest value appears around 15:00 in general.（2）In Wuxi area,the visibility was low in the South part and North part but high in the middle part,and the annual,seasonal and monthly average visibility values gradually decreased from the Wuxi Municipal area to the outside.The occurrence frequency of low visibility weather in Wuxi expressway network presented the distribution characteristics of high in the South and North parts but low in the middle part.（3）Among the meteorological factors affecting the visibility change,the relative humidity was highly negatively correlated with the visibility.The instantaneous wind speed,2-minute average wind speed and the 10-minute average wind speed were highly positively correlated with the visibility,of which 10-minute average wind speed has the greatest positive impact.The correlation between the temperature and the visibility was small,and was greatly affected by seasons.The correlation between the temperature and the visibility was strong positive in summer,weak positive in spring and autumn,but not obvious in winter.（4）Using the above observed meteorological data and reanalyzed data,the data quality was controlled through a K clustering algorithm and a K nearest neighbor algorithm.The data after the quality controlling was normalized by using the wind direction normalization method of wind speed and direction and the minimum and maximum normalization method of other meteorological elements.The processed data was reconstructed according to the idea of time series analysis and supervised learning,The visibility prediction model based on bidirectional long-short term memory and other visibility prediction model based on one-dimensional convolutional neural network and the gated cyclic unit were input,and the random descent algorithm of driving mechanism was used to optimize the model.The super-parameters of the model were determined through a set of experiments.The results of test and error analysis indicated that the model had a high prediction accuracy and good stability.