Dynamic Control of Airport Departures: Algorithm Development and Field Evaluation

Abstract

Abstract-Surface congestion leads to significant increases in taxi times and fuel burn at major airports. In this paper, we formulate the airport surface congestion management problem as a dynamic control problem. We address two main challenges: the random delay between actuation (at the gate) and the server being controlled (the runway), and the need to develop control strategies that can be implemented in practice by human air traffic controllers. The second requirement necessitates a strategy that periodically updates the rate that departures pushback from their gates.We model the runway system as a semi-Markov process using surface surveillance data. We use this modeling framework to derive optimal pushback policies to control congestion. Finally, we present the results of the real-world implementation and field testing of this control protocol at Boston Logan International Airport.

Control and optimization algorithms for air transportation systems

“…The policies were shown to be fair, in that for every minute of gate-hold that an airline experiences, it also receives a minute of taxi-out time savings. In addition, the policies were shown to accommodate prac-250 tical constraints, such as gate-use conflicts, when a departure would need to leave the virtual departure queue (its gate) early because the next aircraft to use that gate had arrived [39].…”

“…On-off or event-driven pushback control policies 165 (such as a threshold heuristic) are not desirable in practice, since both air traffic controllers and airlines prefer a pushback rate that is periodically updated. This observation motivates the development of Pushback Rate Control policies, wherein an op-170 timal pushback rate is recommended to air traffic controllers for each 15-minute interval, and the rate is updated periodically [11,39]. The threshold N-Control policy can be adapted to obtain a rate control policy by predicting the average throughput 175 under saturation over the next 15-minute interval, and then determining the number of pushbacks in that interval that would help maintain the desired level of traffic (typically around the threshold value at which the throughput saturates).…”

“…190 trol A better approach to accommodating the uncertainties in throughput and taxi-out times is through the formulation of a dynamic control problem. Using a queuing model of the departure process built 195 from operational data, we can use dynamic programming to determine the optimal pushback rate that minimizes taxi-out times, while still maintaining runway utilization [39]. The dynamic programming formulation consid-200 ers the system state consisting of the number of aircraft taxiing to the runway and the length of the departure runway queue.…”

“…Using predictions of the runway throughput in the next time period, the runway queue is modeled as a semi-Markov process, 205 and the system state is projected by solving the resultant Chapman-Kolmogorov equations. The optimal policy is determined by solving the Bellman equation for the infinite horizon average cost problem [39]. Fig.…”

“…In comparing the resultant policy with the adapted N-Control policy described in Section 2.3.1, the dynamic programming approach is found to handle uncertainty better (as expected), and results in more robust policies that reduce the risk of runway starvation. The implementation of the dynamic programming based pushback rate control policies at BOS in 2011 showed that during eight 220 4-hour tests, taxi-out time was reduced by nearly 13 hours, while fuel use was reduced by more that 8,200 kg [39].…”