In this paper we use the stochastic branch and bound method to solve the runway scheduling problem with uncertain
input parameters. More specifically, we assume that the time by which an aircraft is available for scheduling is given in
the form of a probability distribution. We consider a runway configuration with two parallel runways, where arriving aircraft
need to cross the departure runway. Given an input schedule and specified distributions describing the uncertainty, the algorithm
generates a sequence of aircraft operations on the runway that minimize the expected makespan.
In order to obtain solutions in a short amount of time we propose several enhancements to the stochastic branch and
bound algorithm. By dynamically changing the number of samples used to estimate the upper and lower bounds during
the course of the algorithm, we can place less emphasis on parts of the branch and bound tree that are unlikely to contain
good solution. Furthermore, we ensure that the algorithm always terminates with the best solution obtained so far, even if
we have not found a complete sequence. With these enhancements, we are able to obtain high quality solutions using less
than 1 min of computation time. More specifically, for instances with 14 aircraft or less, the average runtime for the algorithm
is less than 1 min.
The computational results indicate that the make span decreases by 5–7% when using the stochastic branch and bound
algorithm as compared to an aircraft sequence obtained from a deterministic model. This represents a significant saving
if the result can be translated to a longer planning horizon. In this work we assume that the deviation from earliest runway
time is independent between aircraft, whereas in reality there are several cases where delay is dependent. As an example,
consider a sequence of departing aircraft taxiing along the same taxiway. If the first aircraft is delayed, the remaining aircraft
are likely to be delayed as well. The impact of dependent aircraft operations is a direction of future research.
One major challenge in the stochastic branch and bound algorithm is to estimate good lower bounds in a short amount of
time. In this application we use a partial enumeration scheme, taking advantage of heuristics incorporating constrained position
shifts. It is not within the scope of this paper to explore the trade-off between runtime and solution quality for different
lower (and upper) bound models, rather we suggest that as another direction of future research.