A Simheuristic Method for Airport Runway Scheduling

Shone, Robert and Glazebrook, Kevin and Zografos, K. G. (2022) A Simheuristic Method for Airport Runway Scheduling. In: The Operational Research Society's Annual Conference, 2022-09-13 - 2022-09-15, University of Warwick.

Abstract

Runway scheduling (also known as aircraft sequencing) involves controlling the sequence of aircraft landings on a runway in order to optimise delay-related performance measures. In practice, air traffic controllers might use a ‘first-come-first-served’ policy so that aircraft land in the same order that they arrive in the terminal area, but this is not always the most efficient approach, as the separation requirements between consecutive aircraft pairs also depend on aircraft weight classes – with larger gaps usually required if the leading aircraft is in the ‘heavy’ class. In the academic literature, this type of problem has been formulated as a multi-objective combinatorial optimisation problem. Some classical formulations assume that the problem is both static (i.e. the landing sequence only needs to be determined once, without any subsequent updating) and deterministic (i.e. all relevant information is known, without any uncertainty). However, in reality, the problem is both dynamic and stochastic. In this talk we consider a dynamic, stochastic runway scheduling problem in which the system state at any point in time is high-dimensional and includes the latest estimated times of arrival (ETAs) of planes due to land at the airport, the latest positions of aircraft that have already been ‘queued’ for landing, and also the weather conditions and latest forecast. The ETAs and weather forecasts vary according to continuous-time stochastic processes. The problem is too complicated to be solved using exact methods and we therefore introduce a novel simheuristic method, which involves continuously simulating the performances of various possible landing sequences and using a ranking and selection method to update the hypothesised ‘optimal’ sequence. Preliminary results suggest that our simheuristic method can outperform alternative heuristics that use ‘expected value’ estimates based on the latest system information and treat the problem as if it were deterministic.