TY - CONF

T1 - ADP strategies for resource allocation at congested airports

AU - Shone, Robert

AU - Glazebrook, Kevin David

AU - Zografos, Konstantinos

PY - 2018

Y1 - 2018

N2 - In modern transportation systems there exists a need to develop fast, responsive and easily adaptable methods for computing optimal (or near-optimal) solutions to problems in which resources must be allocated dynamically in order to satisfy time-varying demands from multiple sources. In this talk we consider the case of a single airport which, in response to a pre-determined schedule of arrivals and departures, must use its runway capacity efficiently in order to minimise an objective function based on weighted second moments of aircraft queue lengths.In keeping with a well-established convention in the literature, we model departures and arrivals as independent stochastic queues with time-varying arrival and service rates. Service times are assumed to follow Erlang distributions, whereas for the arrival distributions we consider two possible cases: non-homogeneous Poisson processes and pre-scheduled arrivals with random deviations. We discuss how to formulate the problem of optimising airport capacity usage as a Markov decision process (MDP), and introduce a “surrogate problem” which closely resembles our original problem during periods of heavy demand. We then show that, in our surrogate problem, the MDP value function can be represented as a quadratic function of the state variables, and use this principle to develop ADP strategies for optimising capacity utilisation.

AB - In modern transportation systems there exists a need to develop fast, responsive and easily adaptable methods for computing optimal (or near-optimal) solutions to problems in which resources must be allocated dynamically in order to satisfy time-varying demands from multiple sources. In this talk we consider the case of a single airport which, in response to a pre-determined schedule of arrivals and departures, must use its runway capacity efficiently in order to minimise an objective function based on weighted second moments of aircraft queue lengths.In keeping with a well-established convention in the literature, we model departures and arrivals as independent stochastic queues with time-varying arrival and service rates. Service times are assumed to follow Erlang distributions, whereas for the arrival distributions we consider two possible cases: non-homogeneous Poisson processes and pre-scheduled arrivals with random deviations. We discuss how to formulate the problem of optimising airport capacity usage as a Markov decision process (MDP), and introduce a “surrogate problem” which closely resembles our original problem during periods of heavy demand. We then show that, in our surrogate problem, the MDP value function can be represented as a quadratic function of the state variables, and use this principle to develop ADP strategies for optimising capacity utilisation.

KW - Queueing; Approximate dynamic programming; Airport operations

M3 - Abstract

T2 - StochMod 2018

Y2 - 13 June 2018 through 15 June 2018

ER -