Forecasting workload and airspace configuration with neural networks and tree search methods

The aim of the research presented in this paper is to forecast air traffic controller workload and required airspace configuration changes with enough lead time and with a good degree of realism. For this purpose, tree search methods were combined with a neural network. The neural network takes relevant air traffic complexity metrics as input and provides a workload indication (high, normal, or low) for any given air traffic control (ATC) sector. It was trained on historical data, i.e. archived sector operations, considering that ATC sectors made up of several airspace modules are usually split into several smaller sectors when the workload is excessive, or merged with other sectors when the workload is low. The input metrics are computed from the sector geometry and from simulated or real aircraft trajectories. The tree search methods explore all possible combinations of elementary airspace modules in order to build an optimal airspace partition where the workload is balanced as well as possible across the ATC sectors. The results are compared both to the real airspace configurations and to the forecast made by flow management operators in a French en-route air traffic contro