Study On Visibility Forecasting Of Key Areas Based On Spatio-Temporal Features

Low visibility is an important cause of various traffic accidents.Taking air traffic as an example,between 1919 and 2019,flight accidents caused by low visibility accounted for about 21.2% of all accidents.Accurate visibility forecasts for key areas such as airports and docks can effectively reduce accidents caused by low visibility.However,most of the existing visibility forecasting algorithms are aimed at medium and high scale forecasting,and the number of visibility forecasting algorithms for key areas such as airports and docks is currently small and the accuracy is limited.Aiming at this problem,this paper takes Nanjing Lukou International Airport as an example to study the visibility forecasting of key areas based on spatio-temporal features,and specifically carry out the following work:A visibility forecasting algorithm based on O-Informer(Optimized Informer)is proposed(temporal feature).Firstly,LSTM,Bi-LSTM,Transformer,and Informer are used to build benchmarks for algorithm comparison.Secondly,one-dimensional convolution is used to realize the functions of feature integration,dimension transformation,quantity control,etc.,the network structure of Informer is optimized,and the O-Informer visibility forecasting algorithm is proposed.On this basis,the O-Informer algorithm is verified by using the real measured temporal features dataset.The verification results show that "in the visibility forecasting algorithm based on temporal features constructed in this paper,the performance of the O-Informer visibility forecasting algorithm is superior to other algorithms".Finally,taking the O-Informer visibility forecasting algorithm as an example,the visibility forecasting algorithm based on temporal features is summarized and analyzed.The research results show that "the visibility forecasting algorithm based on temporal features only performs well when the total forecast duration is short,and the performance of such algorithms decreases significantly when the total forecast duration increases."A visibility forecasting algorithm based on O-Pi T(Optimized Pi T)is proposed(spatial feature).Firstly,VGG,Res Net,Vi T,and Pi T are used to build benchmarks for algorithm comparison.Secondly,based on the prior knowledge,the pooling layer and the result vector of Pi T are optimized,and the O-Pi T visibility forecasting algorithm is proposed.On this basis,the algorithm is verified by using the real measured spatial features dataset.The verification results show that "in the visibility forecasting algorithm based on spatial features constructed in this paper,the performance of the O-Pi T visibility forecasting algorithm is better than other algorithms".Finally,taking the O-Pi T visibility forecasting algorithm as an example,the visibility forecasting algorithm based on spatial features is summarized and analyzed.The research results show that "the visibility forecasting algorithm based on spatial features can make more accurate forecasts for high-visibility data,but there are still large errors in the forecast of low-visibility data by such algorithms."An encoding fusion visibility forecasting algorithm and a decoding fusion visibility forecasting algorithm are proposed(spatio-temporal feature).Firstly,the encoder and decoder are used to fuse the temporal and spatial features of meteorological factors,and the encoding fusion visibility forecasting algorithm and decoding fusion visibility forecasting algorithm are proposed.Secondly,the two algorithms are verified based on the real measured spatio-temporal features dataset.The verification results show that "the performance of the decoding fusion visibility forecasting algorithm is better than that of the encoding fusion visibility forecasting algorithm".Finally,different types of visibility forecasting algorithms are analyzed.The research results show that "the visibility forecasting algorithm based on spatio-temporal features can effectively make up for the shortcomings of the visibility forecasting algorithm that only uses a single type of feature,that is,this type of algorithm has a longer forecast duration,better forecast performance,and smaller forecast errors for low visibility data."...