Flight Delay Propagation Analysis Based On Improving Bayesian Networks Structure Learning

To control and reduce flight delays is a long-term and major task of civil aviation management both at home and abroad. To analyze flight data, discover their inner features and find out the changing tendency of flight delay and propagation plays a guiding role in researching flight delays.By an integrated quantitative and qualitative analysis, the theoretical problem of flight delay is studied in this dissertation. The Bayesian network theory is applied to establish a model with real flight data, to analyze the cause effect relationship of influencing factors in flight delays and to present the probability distribution of flight delays under different conditions.In chapter 1, the problem of flight delays, the analysis of flight delay and delay propagation, the summary of present research conditions both at home and abroad are introduced. The process of knowledge discovery based on Bayesian networks learning is described and the main research method of this dissertation is presented.In chapter 2, Bayesian networks learning and basic concepts are briefed. Main methods of parameter learning, structure learning, scoring model and model optimization of Bayesian networks are discussed.In chapter 3, the algorithm of high score priority of genetic-simulated annealing (SA)of Bayesian networks structure leaning is proposed. SA search algorithm and genetic algorithm (GA) of the optimization problem are applied into the structure learning to effectively avoid the prematurity of population development brought by high-score individuals misleading and to improve the precision of Bayesian networks structure learning.In chapter 4, the algorithm of Bayesian networks structure learning based on GA & TS (Taboo Search) is put forward. The method of TS algorithm is applied into the Bayesian networks structure learning based on GA to further improve the crossover and mutation processes so as to enhance the efficiency of Bayesian networks structure learning.In chapter 5, the representative data set of datamining is utilized to make experimental analysis on HSPGSA and GATS of improved Bayesian networks structure learning. The comparative studies with the traditional structure learning algorithm are made to explain the effectiveness and superiority of the method.In chapter 6, A Bayesian network model of flight data is established.On the basis of aforesaid theoretical studies, real civil flight data were collected, the departing flight delay model of large hub airport and the sequence flight delay propagation model of large airline companies are composed respectively to make flight delay propagation analyses. The result is helpful for the executive level to make decisions on related problems of flight delays.