Research On Airport Low Visibility Forecast Technology Based On Deep Learning

With the rapid development of my country's civil aviation industry,related issues caused by visibility have gradually attracted people's attention.This paper selects Shuangliu Airport as the research object of visibility to study the characteristics of visibility changes from 2007 to2017.At the same time,it uses relevant data to analyze a typical low-visibility weather process at Shuangliu Airport and analyzes the main factors that dominate the changes in this lowvisibility process in order to explain The reasons for the formation,continuation and dissipation of this low visibility weather also provide a theoretical basis for the visibility forecast in the later period.Finally,based on the Long Short-Term Memory Model(LSTM),two different forecasting methods are used to forecast the visibility of Shuangliu Airport,and the results are compared and analyzed.The main research results of this paper are as follows:(1)Through statistical analysis to study the inter-annual,seasonal,rose and diurnal variation characteristics of visibility at Shuangliu Airport,the results show that the annual average value of visibility increased by 1907.6m from 2007 to 2017;preference for spring and summer seasons,poorer autumn and winter seasons;monthly average The minimum and maximum visibility appear in December and January respectively;14 o'clock is the maximum visibility of the day,and its minimum appears at 0 o'clock;at the same time,the classification study shows that the number of days with visibility below 3000 meters and the occurrence ratio A decreasing trend year by year;Judging from the number of occurrences of weather processes lower than airport takeoff and landing standards,the number of occurrences is decreasing year by year,with a decrease of 50% throughout the process;it has significant seasonality,with the most in autumn and winter,accounting for 83.03%,While only 3.67% in summer.(2)Research on a typical low-visibility weather process at Shuangliu Airport on December7,2016.The results show that the 500 h Pa weather chart has little change.The central and western regions of China are controlled by a weak high-pressure ridge,and the northwest airflow is dominant.The changes in the weather situation have little impact on the generation and dissipation of low-visibility phenomena;the low-visibility weather changes this time The fundamental reason for the short-term and rapid decrease in visibility during the process is that the sinking airflow under the control of the high-pressure ridge is conducive to the cooling of the ground at night,forming a radiation inversion layer;finally,there is a stable high-pressure ridge on the 500 h Pa level,the relative humidity of the ground is close to saturation,and the lower level is large.The long-term existence of the temperature inversion layer due to the blocking effect of fog weather on solar radiation and the combined effect of the special topography of the Sichuan Basin are the fundamental reasons for the long duration of this low visibility process.(3)Research on visibility prediction technology using neural network,the results show.The key factors that have a greater impact on airport visibility are: 10 m wind field,2m relative humidity,geopotential height and mean sea level pressure.The MAE value and R2 for rolling forecast the visibility in the next 3 hours using the model's forecast data combined with the numerical forecast data are 549.526 and 0.9287,respectively,while the combination of historical observation data and numerical forecast data is used to forecast the MAE value and the visibility of the visibility in the next 3 hours.R2 is 519.865 and 0.9392,respectively.The fitting effects of the two forecasting methods are both higher than 0.9 and there is no overfitting phenomenon.The main reason that the forecast result of the second forecast method is better than the forecast result of the first forecast method is: in the first forecast method,as the forecast progresses,the error of using the forecast value for re-forecasting will follow.Increase.As airport operators in first-line applications are more concerned about low-visibility forecasts,this paper randomly selects two low-visibility change processes and conducts a separate analysis.The results show that: for low-visibility forecasts,observation data and numerical forecasting system data are used for 3 hours The forecasting method is better than using forecast data and numerical forecast data for rolling forecast visibility in the next 3 hours,which is manifested in high forecast accuracy,smoother forecast curve and better fitting effect.