Data-based Inference And Risk Recognition For Civil Aviation Unsafe Events

With the level of China’s national economy as well as the rapid development of air transportation, more and more people choose the aircraft to be the way of their journey. Consecutive air disasters in this year also bring people’s attention to the air aviation safety. Previous studies have been carried out to assess and investigate the risk of air aviation safety in terms of human, machine, environment and management. However, research on unsafe events of civil aviation started late. Though traditional methods of data mining could achieve the purpose of feature analysis, it is difficult to infer incomplete unsafe events and identify the high risk of unsafe events with noised data. In order to tackle these two problems, corresponding approaches have been developed in this thesis and verified with related cases.(1) Fluctuation characteristics analysis of civil aviation unsafe eventsCivil aviation unsafe events are classified and described in this section. Then statistical analysis, time series analysis and association rules analysis are implemented to explore the fluctuation characteristics of the aviation unsafe events. Afterwards, through summing up the merits and demerits of these three methods, two fundamental problems are posed to be addressed in this thesis:inference of incomplete unsafe events and high risk identification of noised data(2) Prediction and inference of civil aviation incomplete events based on Bayesian NetworkIn view of the current situation of incomplete data in civil aviation, this paper makes a reasonable inference and prediction for the unknown risk factors based on Bayesian Network. Grouping unsafe events to complete data and incomplete data, a combined approach with hill-climbing search and score function is adopted to construct the optimal network which produces a good fitness with the complete data. The missing causes of civil aviation incomplete events are predicted and inferred with the advantage of clique tree propagation algorithm which ensures the integrity of the data and points out the direction for the aviation accident investigation.(3) Risk identification of civil aviation unsafe events based on CEEMDA new application based on complementary ensemble empirical mode decomposition (CEEMD) is introduced for the first time to remove noise and identify high risk periods of civil aviation unsafe events. With the cases of bird strikes, the periodicity and tendency have been extracted after examining the performance of each intrinsic mode function (IMF) and residue (R) at different time scales. In comparison with the statistical results, it shows a perfect agreement, which demonstrates the effectiveness of this method. In addition, differential operation is introduced to recover the state of initial data and the high risk periods of bird strikes by month are compared with that before denoising.