Research And Implementation Of Flight Support Process Time Prediction Based On Deep Learning

The rapid development of the domestic aviation industry drives the continuous growth of civil aviation transportation business.Therefore,airports are under enormous operating pressure.As the flight support process is an important part of airport operation,dynamical predict of its execution time can meet the needs of refined management of airports and reduce the risk of flight delays.Existing studies on flight support process time prediction are less combined with process-related theories,and the accuracy and flexibility of the prediction need to be improved.Therefore,this thesis combines the related research in the field of predictive process monitoring and deep learning,proposes a method for predicting the remaining execution time of the flight support process,and builds a predictive monitoring system for flight support process based on this method.Main contributions of this thesis are as follows:(1)Convert the flight support process instance data to event logs,and analyze and extract static features and dynamic features from them.Static features are case attributes that have an important impact on process time,and dynamic features are prefixes composed of events.According to the characteristics of the flight support process,an event representation learning method based on the LINE algorithm is proposed.The event representation vector generated by this method retains relationships of restraint between events in the process as prior information,and outperforms OneHot encoding and Word2 Vec algorithm in the remaining execution time task of flight support process.(2)Considering the differences and correlations between prefixes of different lengths,an iterative training strategy based on transfer learning is proposed.By introducing the attention mechanism,bidirectional structure and a feature fusion module into the LSTM network,the SFF-AttBi LSTM model was constructed to predict the remaining execution time of flight support process.The static and dynamic features extracted from the process instance are the input of the model.The prediction accuracy of the model is better than the LSTM and the annotated transition systems in the task of predicting the remaining execution time of flight support process.(3)Based on the above research,a predictive monitoring system of flight support process is designed and implemented,which has realized the monitoring of operation nodes and the dynamic prediction of execution time of the flight support process being executed.The system also implements functions such as event log management and predictive model configuration.