Using AI (Pytorch) to Optimize Top EU Airport Post-covid Re-opening
Among all economic sectors impacted by Covid19, air traffic was hit the hardest. Airlines are suffering a lot, and consequently so do airports. Now that most EU countries are reopening their borders, air traffic is slowly going up again. Obviously, current traffic is nowhere near pre-covid levels, and it does not make sense for an airport to open all its terminals. The stakes are high: delaying a terminal reopening by even a few months can represent dozens of millions of euros of savings. I was recently involved in a project with #1 EU airport to work on this problem: if you’re an airport, given airline traffic forecasts over the next decade, and financial data on your terminals (fixed/variable costs/revenue), how can you optimize the reopening sequences for all terminals?
First of all, you have to define the objective. In our case, the objective was to maximize profit, and minimize capacity mismatch (between airlines traffic forecasts and open terminals capacity), which is reasonable. Concretely, my plan was to load traffic forecasts and financial data into Pytorch tensors, and then use Autograd to run an optimization with a carefully-built loss function (with a profit term, and a capacity mismatch term).
