Innovative Applications of O.R.Data and queueing analysis of a Japanese air-traffic flow
Introduction
Many airports and airspaces are already congested today, but growing traffic demand will cause increasing delays in the future if no actions are taken (ICAO, 2005, JPDO, 2010, Sesar-Consortium, 2007). Congestion delays materialize either on the ground, where aircraft have to wait before accessing a runway or during the flight, where they are deviated from their intended trajectory. When the schedule buffers are tight, delays may also propagate through the whole transportation network. Congestion delays can be managed on a strategical level (by runway expansion or shorter separation standards), a pre-tactical level (by splitting flows and sectors) or a tactical level (by sequencing and re-sequencing aircraft).
Fig. 1 shows the typical situation. Prior to take-off, departure slots are allocated in order to balance demand with available capacity. Before landing, flows are merged and sequenced in order to efficiently use the runway. In this context, one speaks of radar vectoring, when traffic controllers ask the pilots to stretch their flight paths in the form of specific flight angles and of holding stacks, when pilots are asked to circulate in the vicinity of the airport. Departure slot allocation and en-route sequencing operations create delays, and we call them metering delays in this paper.
The Japanese flow managers currently allocate up to 10 min of metering delay to the en-route phase and the remaining delay to the ground. One can argue that en-route delays are unnecessary, once ground delays are correctly allocated. At least, speed control shall synchronize traffic flows so that radar vectoring can be reduced. But due to arrival-time uncertainties, en-route delays serve as buffers in order to keep a high pressure on the runways. On the other hand, future reduction of arrival time errors promises to reduce en-route delays, allocating more and more delay on the ground. The benefits are smoother traffic flows, less controller workload and reduced fuel consumption. In this paper we derive a decision rule to quantify this new balance.
The deterministic ground delay and aircraft sequencing problems are well known in the OR and the transportation communities, see for example Erzberger, 1995, Beasley et al., 2000, Bayen et al., 2004, Lulli and Odoni, 2007, Artiouchine et al., 2008, Balakrishnan and Chandran, 2010. The stochastic phenomenon of congestion was pioneered by Cox (1955), defining it as
(i) a flow of customers needing service,
(ii) some restrictions on the availability of service, and
(iii) irregularity in the flow of customers, the servicing operation or both.
Moreover, when average demand is higher than available capacity (e.g. during peak hours or bad weather conditions), predictable congestion occurs. Phenomena with the above characteristics, such as telecommunication networks, passenger flows, and many more, are often modeled as stochastic or deterministic queueing systems (Newell, 1982, Wolff, 1989). The air transportation system also exhibits the classical queuing behavior: as demand approaches capacity, delays increase sharply (Ball et al., 2010).
Early contributions in this regard are by Bell, 1949, Dunlay and Horonjeff, 1976. Often, arguments for Poissonian arrival flows are elaborated on which steady-state control strategies are based. A homogeneous Poisson process assumes essentially that the numbers of arrivals in disjoint time intervals are independent random variables, and that no more than one arrival takes place per time instant (see Daley and Vere-Jones (2003) for an axiomatic derivation and historic derivations from Erlang and Bateman). Also since long time, the existence of steady-states in air transportation are questioned (e.g. Barnhart et al., 2003, Janic, 2000). One of the few transient analyzes is Peterson et al. (1995), based on an algorithmic approach. We will later show that the Poisson hypothesis is not unreasonable whenever congestion is moderate, and that steady-states can be reached for pre-scheduled flows in the order of a few hours.
The assumptions of homogeneous Poisson flows have been relaxed in at least three ways. Non-homogeneous queueing models allow for variations of arrival and service rates. This seems more realistic to capture high traffic peaks, particularly during bad weather. Typically, time-dependent Poisson arrivals are modeled with time-dependent service distributions (Bäuerle et al., 2007, Fan and Odoni, 2002, Hansen et al., 2009, Knessl and Yang, 2002, Pyrgiotis et al., 2013, Yang and Knessl, 1997). Models with deterministic service times serve as lower bounds and models with exponentially distributed service times as upper bounds to the real delays (Barnhart et al., 2003). Multiple arrivals per time instant (batch arrival models) are another relaxation of the Poisson assumption. Corresponding queueing models are proposed for aircraft landings (Falin, 2010, Krishnamoorthy et al., 2009). More recently, a third departure from Poisson arrivals was presented in Guadagni et al., 2011, Guadagni et al., 2013. They clarify that the numbers of arrivals in subsequent time intervals will be negatively correlated, whenever aircraft are pre-scheduled at a critical resource but arrive with random errors. Pre-scheduled queues are also studied in the field of ‘appointment scheduling’, where customers receive initial assignments, but may arrive with a positive or negative delay. From the queueing methodology point of view, there was some activity in the 1960’s and 1970’s, but it loses its traces then (Bloomfield and Cox, 1972, Mercer, 1960, Mercer, 1973, Winsten, 1959). Today, the field is widely occupied with algorithmic questions (e.g. Begen and Queyranne, 2011, Green, 2008 and the references therein) or numerical approaches (Nikoleris and Hansen, 2012).
The remainder of the article is organized as follows: in the next section we describe a radar-data analysis of arrival traffic for a Japanese airport. We then analyze two types of queueing models as generators for metering delays: pre-scheduled random arrivals (PSRA) and Poissonian arrivals. Based on this, we derive a rule how to balance metering delays between ground and air efficiently.
Section snippets
Data analysis
We analyzed available radar data in order to get a first impression on the statistical properties of the traffic flows and the generated delays. We selected nine days of ‘normal’ traffic from the months August, October and December 2008, i.e., where no exceptional events or delays were reported. We removed outliers by hand (about 10% of missing or erroneous fields in the source data). As main results we found empirical evidence for and against a Poisson hypothesis during moderate congestion and
Queueing analysis
The empirical analysis confirmed that demand fluctuations are the major driver for metering delays, so we modeled the metering gate as a single server queueing system. Input to the system are flows from the different routes. Output is a single flow, separated by at least = 10 NM. Depending on speed, wind conditions and human factors, it may vary from time to time. We expected from the analysis to further clarify which part of the observed delays is due to metering and which to operational
Queueing results
We fitted the various queueing models on our instance. The major questions were (i) the transient behavior of the PSRA model, (ii) the difference between PSRA and Poissonian models, (iii) the difference between deterministic and general service times and (iii) delay patterns that cannot be explained by queueing models.
Conclusions
In this paper we analyzed the most congested traffic flow in Japanese airspace. The current strategy balances delays between the ground and en-route in order to take into account arrival time errors. As long-term strategy, flow managers wish to reduce the amount of en-route delay in order to smoothen the flows. Our approach was empirical, supported by standard and non-classical stochastic queuing models.
A radar-data analysis revealed ‘almost’ Poissonian flow characteristics in arrival rate and
Acknowledgments
This work was partly supported by the Electronic Navigation Research Institute (Japan) under the study on Trajectory Modeling and Prediction. The authors are grateful to four anonymous referees for their careful reports and to Masato Fujita for comments on a draft of this paper.
References (45)
- et al.
Runway sequencing with holding patterns
European Journal of Operational Research
(2008) - et al.
Algorithms for scheduling runway operations under constrained position shifting
Operations Research
(2010) - Ball, M., Barnhart, C., Dresner, M., Hansen, M., Neels, K., Odoni, A., et al. (2010). Total delay impact study....
- et al.
Applications of operations research in the air transport industry
Transportation Science
(2003) - et al.
On the waiting time of arriving aircrafts and the capacity of airports with one or two runways
European Journal of Operational Research
(2007) - et al.
An approximation algorithm for scheduling aircraft with holding time
Proceedings of the 43rd IEEE Conference on Decision and Control
(2004) - et al.
Scheduling aircraft landings – the static case
Transportation Science
(2000) - et al.
Appointment scheduling with discrete random durations
Mathematics of Operations Research
(2011) Operational research into air traffic control
Journal of the Royal Aeronautical Society
(1949)- et al.
A low traffic approximation in queues
Journal of Applied Probability
(1972)
The statistical analysis of congestion
Journal of the Royal Statistical Society. Series A
Stochastic properties of enroute air traffic – an empirical investigation
Journal of Aircraft
Probability: Theory and examples
A single-server batch arrival queue with returning customers
European Journal of Operational Research
A practical perspective on airport demand management
Air Traffic Control Quarterly
Reducing delays for medical appointments: A queueing approach
Operations Research
Queueing systems with pre-scheduled random arrivals
Mathematical Methods of Operations Research
Cited by (21)
Modelling and assessment of the arrival and departure process at the terminal area: A case study of Chennai international airport
2023, Physica A: Statistical Mechanics and its ApplicationsCooperative surveillance systems and digital-technology enabler for a real-time standard terminal arrival schedule displacement
2021, Advanced Engineering InformaticsApplications of stochastic modeling in air traffic management: Methods, challenges and opportunities for solving air traffic problems under uncertainty
2021, European Journal of Operational ResearchPredictive modeling of inbound demand at major European airports with Poisson and Pre-Scheduled Random Arrivals
2020, European Journal of Operational ResearchCitation Excerpt :In fact, the demand over two consecutive intervals is characterized by non-zero correlations (point estimates mostly negative and often significant at the 5% level), which is not consistent with the assumption of independent increments. Homogeneous or not, Poisson processes remain a popular choice for modeling inbound air traffic (Gwiggner & Nagaoka, 2014). Thus, we have proposed a novel data-driven approach to the modeling of a non-homogeneous Poisson process by use of pelt (a change-point detection algorithm) and dbscan (clustering).
Resource allocation in congested queueing systems with time-varying demand: An application to airport operations
2019, European Journal of Operational ResearchCitation Excerpt :A key factor behind this trend has been the development of projects such as NextGen in the USA (JPDO, 2010) and SESAR in Europe (European Commission, 2018), which are intended to make flight arrival times more predictable through the use of four-dimensional trajectory-based operations. We therefore include Case 2, which bears similarities to pre-scheduled random arrival (PSRA) models considered in recent literature (Caccavale, Iovanella, Lancia, Lulli, & Scoppola, 2014; Guadagni, Lancia, Ndreca, & Scoppola, 2013; Gwiggner & Nagaoka, 2014; Nikoleris & Hansen, 2012), as an alternative to the Poisson model. In fact, it can be assumed that the decision-maker always chooses a point on the upper boundary of Rs, since the objectives that we consider in this paper imply that one should always choose the largest possible value of νs for a given value of μs (this can be shown using a simple sample path argument, which we have omitted here).
Minimizing deviation from scheduled times in a single mixed-operation runway
2017, Computers and Operations ResearchCitation Excerpt :In this paper, we will evaluate a simulated annealing (SA) algorithm designed to calculate the landing and/or departure times, minimizing the total delay from the estimated landing time (ELDT) or from the estimated time of departure (ETD), subject to wake vortex separation (WVS) requirements and constrained position shifting (CPS). Following the Gwiggner and Nagaoka [27] classification, this research is placed at the tactical level and focused on airports with single, mixed-operation runways that need to optimize their resources to deal with capacity problems. We have considered three factors of interest for the algorithm evaluation: number of aircraft, wake-turbulence category and number of shifts of the aircraft from their initial position in the sequence.