Future Air Routes Discovery Based On Link Prediction

Analyzing the evolution of air transport network structure and predicting the most likely new air routes in advance is not only an important basis for civil aviation departments and local governments to adjust their development strategies in time,but also of great significance to enhance the robustness of network structure and ensure the sustainable development of the entire network.At present,the link prediction technology is still in the exploratory stage in the future air routes discovery,and it mainly adopts the similarity-based methods,which have some problems such as subjective selection of similarity indicators and insufficient network information mining.In order to solve these problems and further improve the prediction accuracy as well as model robustness,graph neural network and link prediction technology are combined to research the future air routes discovery on the actual operation data of Chinese air transport network.The specific research work are as follows:In view of the similarity-based link prediction methods,the indicators selection relies on artificial assumptions and the poor scalability of the model.Starting from the overall structure of air transport network,extract the local enclosing subgraphs,label the structural roles of nodes,and then use the self-learning characteristics of graph convolutional neural network to learn the reasons and driving forces of new air routes,and build a link prediction model NARP(New Air Routes Prediction)based on graph convolutional neural network to realize future air routes discovery.The experimental results on the actual operating data of Chinese air transport network show that compared with the benchmark algorithm,the prediction accuracy of NARP is up to 9.28% higher,but the robust performance of the model needs to be improved.In view of the single index(only network structure characteristics)and low robustness of the NARP model,considering the political and economic environment of navigable cities and the operation and development status of local airports will affect new air routes,factor analysis and hierarchical aggregation are adopted to extract the hierarchical attributes of nodes(navigable cities),further optimizes the NARP model,and builds a link prediction model NARP-SA(New Air Routes Prediction – Structure & Attributes)that integrates network structure and node attributes.Experimental results show that in the case of extremely incomplete network,the prediction accuracy of NARP-SA can be maintained at about 80%,and the prediction accuracy and robust performance are further improved compared with NARP.Almost half of the top 15 predictions of NARP-SA actually appear in the real Chinese air transport network,indicating that the NARP-SA prediction results are in line with the actual evolution of the network.After the NARP-SA prediction results are added to the air transport network,the connectivity and robustness of the network are greatly improved,which shows that NARP-SA can achieve the goal of network structure optimization on the basis of conforming to the evolution of the air transport network itself.