Predicting Airport Capacities Using Neural Networks

In the domain of air traffic management, reliance on heuristic guidelines for determining airport capacity often leads to suboptimal Traffic Management Initiatives (TMIs). To address this issue, we propose a data-centric methodology that leverages feed forward and recurrent neural networks. These networks analyze diverse inputs including weather forecasts, scheduled demand, time of day, and current airport capacity data to make accurate predictions of future airport capacities and runway configurations. Our model integrates comprehensive datasets, including the High-Resolution Rapid Refresh model and Terminal Aerodrome Forecast weather forecasts, and 2018 to 2019 historical data from Newark Liberty International Airport (EWR). By capturing temporal dependencies in weather forecasts, the neural network offers deterministic predictions of airport capacity and runway configurations. We compare the models' performance in various weather scenarios to provide insights into the practical applications of our data-driven approach. By integrating advanced weather models, historical data, and error assessment, we aim to offer a more accurate and reliable alternative to the current heuristic guidelines, thereby contributing to the growing body of research on data-centric approaches in air traffic management. This holistic approach holds promise for optimizing airport capacity management, predicting runway configurations, and improving overall air traffic flow efficiency.