Benjamin S. Levy

Estimating Taxi-out times with a reinforcement learning algorithm

Flight delays have a significant impact on the nationpsilas economy. Taxi-out delays in particular constitute a significant portion of the block time of a flight. In the future, it can be expected that accurate predictions of dasiawheels-offpsila time may be used in determining whether an aircraft can meet its allocated slot time, thereby fitting into an en-route traffic flow. Without an accurate taxi-out time prediction for departures, there is no way to effectively manage fuel consumption, emissions, or cost. Dynamically changing operations at the airport makes it difficult to accurately predict taxi-out time. This paper describes a method for estimating average taxi-out times at the airport in 15 minute intervals of the day and at least 15 minutes in advance of aircraft scheduled gate push-back time. A probabilistic framework of stochastic dynamic programming with a learning-based solution strategy called Reinforcement Learning (RL) has been applied. Historic data from the Federal Aviation Administrationpsilas (FAA) Aviation System Performance Metrics (ASPM) database were used to train and test the algorithm. The algorithm was tested on John F. Kennedy International airport (JFK), one of the busiest, challenging, and difficult to predict airports in the United States that significantly influences operations across the entire National Airspace System (NAS). Due to the nature of departure operations at JFK the prediction accuracy of the algorithm for a given day was analyzed in two separate time periods (1) before 4:00 P.M and (2) after 4:00 P.M. On an average across 15 days, the predicted average taxi-out times matched the actual average taxi-out times within plusmn5 minutes for about 65 % of the time (for the period before 4:00 P.M) and 53 % of the time (for the period after 4:00 P.M). The prediction accuracy over the entire day within plusmn5 minutes range of accuracy was about 60 %. Further, application of the RL algorithm to estimate taxi-out times at airports with multi-dependent static surface surveillance data will likely improve the accuracy of prediction. The implications of these results for airline operations and network flow planning are discussed.

Assessing the impacts of the JFK Ground Management Program

The Ground Management Program at John F. Kennedy International Airport (JFK) aims to leverage the availability of comprehensive airport surface surveillance data and airline schedule information to better manage the taxi-out process, reduce taxi times, and improve efficiency. During periods when departure demand exceeds capacity, departing aircraft are held at the gate or another holding location, and released to the runway in time to join a short departure queue before taking off. As a result, aircraft absorb delay with engines off, and decrease their fuel burn, emissions, and engine maintenance costs. This paper evaluates data from before and after departure metering was initiated at JFK, to assess its impacts. The results show that airport performance has improved, and that the departure metering is responsible for a significant portion of the improvements. The paper also finds that the new, more automated, Ground Management Program that was implemented in April 2012 has continued to yield significant benefits. The average taxi-out time savings at JFK due to departure metering in the summer of 2012 is estimated to be about 1.5-2.7 minutes per flight.

DEPARTURE TAXI TIME PREDICTIONS USING ASDE-X SURVEILLANCE DATA

Accurate prediction of departure taxi times will help airlines to proactively manage push-times, to optimally assign scarce ramp resources, and to propagate delay information to destination airports in a more timely fashion. Air Traffic Control (ATC) will benefit via improved demand forecasts for the terminal area and enroute air sectors. An ancillary benefit to such predictions is the ability to discern factors contributing to longer taxi times. To facilitate accurate predictions, we will analyze the utility of the Airport Surface Detection Equipment, Model X (ASDE-X) surface surveillance data. Months of archived data support both historical analysis (i.e., under similar conditions, what happened in the past?) and a more dynamic real-time surface analysis (e.g., aircraft positions, queues, and runway utilization).

Benefits Assessment of a Surface Traffic Management Concept at a Capacity-Constrained Airport

Inefficient surface traffic management may lead to congested taxiways, long departure queues, and excess delay in the air transportation system. To address this problem, NASA researchers have developed optimization algorithms and a concept of operations for an airport surface traffic management tool called the Spot and Runway Departure Advisor (SARDA). Past SARDA research efforts have been focused on the Dallas/Fort Worth International airport. This paper describes the development of SARDA-like schedulers for managing the traffic at an operationally dissimilar airport―Charlotte Douglas International airport, and presents the results of a fast-time simulation-based benefits assessment. Fasttime simulations were conducted to test the benefits of optimized scheduling over a baseline model of current-day operations. In the fast-time simulations, it was observed that optimization schedulers reduced movement area delays by up to 3.1 minutes per departure on average, as compared to the baseline simulation. The movement area delay savings translated to shorter movement area taxi-out times and an average reduction in fuel burn and emissions of approximately 24% per departure. The overall trend observed in the total delay (gate delay + ramp delay + movement area delay) comparison indicated the optimization schedulers were not able to reduce total delay, and runway throughput comparisons suggested the optimization schedulers had little to no effect on throughput.

Departure queue management benefits across many airports

Departure queue management holds the promise of improving runway throughput and reducing queue length, taxi time, fuel burn, and emissions, while reducing costs and improving the passenger experience. These benefits are achieved by allowing departures to accept delay at the gate, rather than in a long departure queue, while maintaining their position in a virtual queue. Previous work has evaluated these savings for individual airports, using simulation and data analysis. This paper extends the previous work to assess the potential impact of a queue management policy across a large number of airports. One factor often neglected in previous studies is the impact of limited gate capacity, which could limit the achievable savings at some airports. We study the effect of departure queue management on gate occupancy, and find that limited capacity imposes only a modest reduction in savings. We also evaluate the economic value of the savings offered by queue management, and find that they amount to tens of millions of dollars per year at a large airport, even before considering additional benefits such as prioritization of high-value flights and the reduction in missed passenger connections. Finally, we compare the potential benefits of departure queue management across 22 U.S. airports, using recorded surveillance data over a long period of time, covering a variety of conditions. The potential benefits of queue management will be greatest at airports where long departure queues, and long taxi-out times, are often observed. Using measurements of queue length derived from surveillance data, we quantify the benefits that may be achieved at each airport and identify those with the greatest potential savings.

Objective and automatic estimation of excess taxi-times

This paper describes the methodology developed at Sensis Corporation for the automatic and objective estimation of total and excess taxi-times from Airport Surface Detection Equipment - Model X (ASDE-X) surveillance data, such that these quantities can be conditioned on the basis of runway and gate/ramp locations. For each airport in the daily summary, we report the number of arrival and departure operations, total taxi-time, excess taxi-time, percent of known aircraft types, and the percent of complete aircraft taxi trajectories. Other data columns in the daily summary include fuel burn, fuel cost, and emissions (i.e., HC, CO, NOx), reported as total and excess quantities. A daily report is automatically generated for the airports at which Sensis Corporation currently makes recordings: ATL, BDL, CLT, DTW, IAD, MCO, MEM, MKE, ORD, PVD, SDF, SEA, and STL; this list will grow as more ASDE-X systems are fielded. Estimation of excess fuel burn and cost requires data on the aircraft type and excess taxi-time. The aircraft type determines the fuel burn rate, taken from the ICAO database; the excess taxi-time depends on a complete taxi trajectory in the movement area. The percent of known fuel burn rates ranges from 85 to 94% for the current set of airports. The percent of complete trajectories ranges from 83 to 93% for taxiing in the movement area. For validation, we have undertaken comparison of operation counts from the processing of ASDE-X data with data reported in the FAAs Aviation System Performance Metrics (ASPM) database, and have found good agreement (standard error < 1 operation). Also, we have performed some comparisons of the ASDE-X total-time estimates against the reportable quantities from the on-time performance database of the Department of Transportation (DOT) Bureau of Transportation Statistics (BTS). This analysis is performed on a per-aircraft basis by matching the tail numbers and out-off-on-in (OOOI time) events between the two data sets.

Arrival Time Estimation (ETA) from On-Final to Gate

[Abstract] Accurate prediction of taxi-times on the airport surface is a key component of the collaborative decision making (CDM) concept of operations for efficient management of airport resources. The benefits of improved accuracy in predicted taxi-times are better gate management and reduced arrival and departure delays. This research supports the development of a predictive tool for the estimation of the arrival time at the gate for an aircraft when it is on its final approach, including estimation of threshold crossing time and on-ground taxi-time. Initial model development for Detroit Wayne County International (DTW) Airport predicts taxi-time from taxi-paths and conditioning factors identified from analysis of surface surveillance data. This work demonstrates the feasibility of automatic identification of taxi-paths from surveillance data, with application to taxi-route conformance monitoring. Development of a rule set for selection of arrival taxipaths (wheels-down to ramp area) shows that it is possible to predict taxi-time with a standard error of 1.67 minutes, based on eight dates of data in May 2006. Conditioning factors used to develop the rule set include arrival runway, ramp destination, aircraft type, and taxi-path.

A Real-Time ETA-to-Threshold Prediction Tool

This paper presents the results from the development of a procedure and algorithms for the real-time estimation of aircraft arrival times (ETA-aloft) at the runway threshold. The prototype was developed with extended range (90-110 nmi) multilateration data measured at Lambert-St. Louis International (STL) Airport. Model development and calibration was based on data collected in April 2005; validation was performed with data collected in February 2005. The model is based on airspace region characteristics, is analytical in nature, and is based on commonly-available data (e.g., aircraft position, runway properties, navaids)

Departure management: Savings in taxi time, fuel burn, and emissions

Departure management holds the promise of improved runway throughput and reduced queue length, taxi time, fuel burn, and emissions. A departure management tool (DMAN) in development at Sensis Corporation achieves these benefits by controlling the times at which aircraft push back from the gate or enter the airport movement area. DMAN automatically determines times for taxi clearance and take-off for each flight, and allows users to modify this schedule as desired. This tool integrates with existing information sources and other decision support tools, requiring minimal equipment investment and minimal changes to operational practice. Therefore, the efficiency benefits of departure management are achievable today. We evaluate the likely benefits of using a departure management tool by performing day-long simulations of traffic at two airports - John F. Kennedy International Airport (JFK) and Philadelphia International Airport (PHL). For each airport, we simulate two scenarios. The first is a baseline in which departures taxi and queue at the runway on a first-come-first-served (FCFS) basis, corresponding to airport operations today. The quantitative accuracy of this model is validated by comparing to recorded surveillance data. In the second simulated scenario, DMAN is used to hold aircraft at the gate and to adjust the departure sequence. Comparing taxi times, fuel burn, emissions, and overall delay between the two scenarios, we find substantial improvement in all of these measures when the DMAN tool is in use.

Analysis and causality of airport surface delays

A prototype departure advisor is under development at the Sensis Corporation, with support from the New York State Energy Research and Development Authority (NYSERDA). The prototype is being developed based on operations data at John F. Kennedy International (JFK) Airport, and depends on surface surveillance data (ramp area, movement area), flight plans, and traffic flow constraints. The prototype will provide recommended departure pushback times for airline ramp managers, and will deliver departure sequences to the movement area in an improved order. Use of the prototype will reduce taxi-out times, fuel burned, and emissions, and improve efficiency of departure runway use. Measurement of the location, duration, and cause of holding on the airport surface by taxiing aircraft is an important part of demonstrating potential benefits from a departure advisor (e.g., reduction in excess fuel burn) and also helps with the development of the prototype itself. As mentioned above, one outcome of the prototype is improved departure sequences, with optimized groups of departures bounded by successive arrivals. Hold categories and algorithms for detection and classification have been developed for arrivals and departures taxiing on the airport surface. Hold categories and algorithms depend on having high-quality surveillance data. Some hold classifications require gate-area surveillance data, such as is available at JFK. The arrival hold categories are: arrival held short of crossing runway and arrival held waiting for departure to clear ramp alleyway. Departure hold categories are: departure queue hold, departure runway pre-roll hold, hold at gate, hold at pushback, and hold at movement area/ramp area ”spot.” Hold categories not specific to operation type (e.g., arrival) are: operation held behind arrival holding short of crossing runway, operation held to merge with or follow taxiing traffic, and operation held to yield to crossing traffic. This paper presents results from the application of the holding algorithms. For example, the material presents statistics on the holding of arrivals short of an active departure runway. Also, examples of departure queue statistics (maximum departure queue depth vs. service time, fraction of time spent in a held position while in departure queue) will be presented. This work extends the state-of-the-art for detection and quantification of aircraft taxi delays on the airport surface by providing some causal attribution (e.g., blocked ramp alleyway). Causality is important to business cases made to support adoption of airport automation technology by an air navigation service provider (ANSP), airline, or airport authority.