Predicting Expect Departure Clearance Times Based on Surface Weather Observations for a Major Hub Airport: A Machine Learning Approach
Presenter Email
misras@my.erau.edu
Keywords
Expect Departure Clearance Time, Machine Learning
Abstract
Commercial air travel in the United States has grown significantly in the past decade. While the reasons for air traffic delays can vary, weather is the largest cause of flight cancellations and delays in the United States. Air Traffic Control centers utilize Traffic Management Initiatives such as Ground Stop and Estimate Departure Clearance Times (EDCT) to manage traffic into and out of affected airports. Airline dispatchers and pilots monitor EDCTs to adjust flight blocks and flight schedules to reduce the impact on the airline’s operating network. The use of time-series machine learning models has demonstrated effectiveness in predicting different types of flight delays. For the purpose of predicting EDCTs based on surface weather observations at Charlotte Douglas International Airport, Vector Autoregression and Recurrent Neural Network, specifically Long Short Term Memory, models were developed. The two models were evaluated on Mean Squared Error, Mean Absolute Error, and Root Mean Squared Error. While both models were assessed to have demonstrated acceptable performance, the Vector Autoregression outperformed the Recurrent Neural Network model. Weather-related variables up to six hours before the prediction time period were used to develop the lasso regularized Vector Autoregression equation. Precipitation values were assessed to be the most significant for EDCT prediction by the Vector Autoregression and Recurrent Neural Network models.
Predicting Expect Departure Clearance Times Based on Surface Weather Observations for a Major Hub Airport: A Machine Learning Approach
Commercial air travel in the United States has grown significantly in the past decade. While the reasons for air traffic delays can vary, weather is the largest cause of flight cancellations and delays in the United States. Air Traffic Control centers utilize Traffic Management Initiatives such as Ground Stop and Estimate Departure Clearance Times (EDCT) to manage traffic into and out of affected airports. Airline dispatchers and pilots monitor EDCTs to adjust flight blocks and flight schedules to reduce the impact on the airline’s operating network. The use of time-series machine learning models has demonstrated effectiveness in predicting different types of flight delays. For the purpose of predicting EDCTs based on surface weather observations at Charlotte Douglas International Airport, Vector Autoregression and Recurrent Neural Network, specifically Long Short Term Memory, models were developed. The two models were evaluated on Mean Squared Error, Mean Absolute Error, and Root Mean Squared Error. While both models were assessed to have demonstrated acceptable performance, the Vector Autoregression outperformed the Recurrent Neural Network model. Weather-related variables up to six hours before the prediction time period were used to develop the lasso regularized Vector Autoregression equation. Precipitation values were assessed to be the most significant for EDCT prediction by the Vector Autoregression and Recurrent Neural Network models.
Comments
Presented in Session 3 A - Advancing Aviation: AI & Machine Learning