Enhancing a simulation-based Allocator Software with machine learning and genetic algorithms for improving the gate assignment problem

Assigning gates to flights considering physical, operational, and temporal constraints is known as the Gate Assignment Problem. This article proposes the novelty of coupling a commercial stand and gate allocation software with an off-the-grid optimization algorithm.

The software provides the assignment costs, verifies constraints and restrictions of an airport, and provides an initial allocation solution. The gate assignment problem was solved using a genetic algorithm. To improve the robustness of the allocation results, delays and early arrivals are predicted using a random forest regressor, a machine learning technique and in turn they are considered by the optimization algorithm. Weather data and schedules were obtained from Zurich International Airport. Results showed that the combination of the techniques result in more efficient and robust solutions with higher degree of applicability than the one possible with the sole use of them independently.