OR/MS Today - April 2003|
Right on Queue
OR models improve passenger flows and customer service at Vancouver International Airport
By Derek Atkins, Mehmet A. Begen, Bailey Kluczny, Anita Parkinson and Martin L. Puterman
As operations research professionals we spend a considerable amount of time traveling to client sites, academic meetings and university seminars. After Sept. 11, 2001, these trips became longer as heightened security measures led to new and more complex security screening processes, sometimes resulting in longer lines and decreased throughput at security checkpoints. Immediately following 9/11, it was not unheard of to spend more time in security lines than in the air.
A timely and innovative OR-based study carried out by Vancouver International Airport Authority (YVRAA) in conjunction with students, staff and faculty from University of British Columbia's (UBC) Commerce's Centre for Operations Excellence (COE) showed that through efficient scheduling and job deployment, 90 percent of Vancouver International Airport (YVR) passengers could expect to wait no longer than 10 minutes at pre-board screening (PBS) security points.
The Vancouver International Airport is operated and managed by the Vancouver International Airport Authority. This community-based, not for-profit organization operates under a long-term lease from the Canadian government. Its focus on safety, security and customer service has contributed to YVR's ranking among the top 10 airports in the world. To maintain its excellent customer service standards and in anticipation of new government regulations, airport management sought to take leadership in improving customer flow through its airport security checkpoints. It was at this point that YVRAA turned to the COE for assistance.
The COE carried out a two-phase study (see the Figure 1 schematic) to address YVRAA managerial needs. The study focussed on the operation of the two domestic, the trans-border and the international pre-board screening locations. In the first phase, the research team developed the YVRAA Security Queuing Simulation Model to compare several operational strategies and determine staff levels required to obtain achievable service standards. The second phase was to generate shift schedules that would achieve these standards at minimal cost.
Figure 1: Schematic overview of study components and outcomes.
While most of us have passed through pre-board screening, it is unlikely that we have paid much attention to the intricacies of this process. With process flow software and stop watches in hand, the research team viewed the pre-board screening operation through analysts' eyes.
Team members (primarily students) spent several days, some starting at 5 a.m., observing the airport's four pre-board security locations. They observed the flow of passengers through the whole screening process, and collected data on passenger characteristics and time spent at each of the individual process steps. This activity alone highlighted some areas for immediate improvement. Recommendations included better signage on payment of the Airport Improvement Fee (AIF), asking passengers to boot up their laptops and place their metal possessions into special containers before entering the pre-boarding screening area, and reconfiguring the layout to allow more space for manual searches. Measures like these immediately cut down on bottlenecks and improved the flow of passengers through the pre-board screening area.
Observations were corroborated by interviews with security personnel. Bringing all the pieces together gave us a thorough understanding of the system's processes and procedures and enabled us to develop preliminary process maps. The preliminary maps underwent a number of revisions to ensure that all parties agreed that they accurately represented system flows and operations.
The Airport Authority provided AIF data from each of the airport's four screening points. AIF data was used as a proxy for passenger volumes at each of the four screening points and was one of the main inputs for the simulation model. It was also used to validate the departing passenger generator used in the staff scheduling analysis described below.
Simulation Development and Benefits
To overcome the complexity of the system, modeling began with only a single screening line. This gave us insight into which aspects of the process we understood well and where we required additional observation and data. Once we were satisfied with our model of a single screening line, we extended it to the full system consisting of five parallel screening lines as depicted in the simulation screen shot (Figure 2).
Figure 2: Screenshot of YVRAA Security Simulation showing the pre-board screening configuration in the domestic terminal. Passenger colors indicate the number of bags carried and security are represented in red when active and green when inactive.
To validate our simulation, we collected additional waiting time and throughput data. We compared the simulation output to this data and revised the logic and service time data until we were satisfied that the simulation output agreed with the observations.
Since pre-board screening operations throughout the airport are similar, further data collection allowed us to extend the model to all four screening areas. Changing parameters such as the bag search ratio and the number of bags carried allowed us to model the international and trans-border screening areas in addition to the domestic screening points. After this step was completed, we were able to conduct meaningful analyses and proceed with the next phases of the project.
The YVRAA Security Queuing Simulation, which was developed in ARENA 6.0, has become a valuable tool to visualize pre-board screening operations identify bottlenecks and conduct "what-if" analyses. Simulation output statistics provide resource utilization, queue lengths and time spent in the system measurements.
Using the YVRAA Security Queuing Simulation, the Airport Authority is able to anticipate the impact of a change in passenger numbers or staffing levels on waiting times. It showed that under some conditions it is more efficient to have two security lines fully staffed than five partially staffed, and that increasing staffing levels could be more effective than acquiring additional expensive machinery. Most importantly, it showed what staffing levels and shift configurations were required to realize the goal of 90 percent of passengers waiting a maximum of 10 minutes in line. YVRAA's Vice President of Operations Craig Richmond noted that "the simulation model helps answer many other questions that may seem simple but can be extremely complex to answer."
Some key observations included:
The second phase of the project sought to determine shift schedules to achieve the 90-10 service criteria with a minimum number of staff hours. Our approach combined a passenger load-forecasting model, simulation to determine staffing requirements across the four PBS lines and linear programming to determine an optimal allocation of shifts. Shift schedules were developed on a daily basis and allowed movement of staff between different PBS locations to take into account the different load patterns.
To be useful by YVRAA analysts, our models had to be easy to understand, quick to execute and compatible with readily available analysis tools. Thus, we focussed on integrating flight schedules, the simulation and the optimization model through Microsoft Excel.
In forecasting passenger demand at each PBS location, the objective was to turn a daily flight schedule listing the departure time, gate number and number of seats for each aircraft, into the estimated number of passengers expected at each PBS location at each 10-minute interval throughout the day. Fortuitously, AIF sales data was available to validate our approach. The AIF is paid by departing passengers at collection points immediately preceding the PBS locations. Comparing this data to hand counts revealed that the AIF data was a good proxy for passenger arrival time data.
We used the flight schedule to determine the capacity and departure time for each flight and estimated the number of passengers on a flight by multiplying its capacity by the anticipated load factor. We then allocated the estimated number of passengers to a time period prior to departure using a triangular distribution. We aggregated data across flights to determine the number of passengers arriving in each time slot. Using the AIF data for validation, different triangular distributions were evaluated until we found a best one for each PBS location. Passenger connection ratios were acquired to determine the approximate connection ratio to use with the passenger generator for the international pier.
We were very pleased with these results; in general, simple triangular distributions produced a demand profile that consistently mirrored most of the peaks and valleys of the AIF data, and was very close in magnitude at the key morning and evening departure times. We settled on separate distributions for flights to Canadian destinations, and flights to the United States and other international destinations. For flights to Canadian destinations, we used a 90-40-20 triangle in which the first passenger arrived at the PBS 90 minutes before departure, the last passenger arrived 20 minutes before departure, and the most likely arrival time at PBS was 40 minutes before departure. For the international and trans-border flights, a 150-80-20 triangle was selected reflecting the earlier suggested check in times for these flights.
Achieving the Service Criterion
The next step was to translate this demand into staffing requirements. We created a look-up table for each PBS location listing expected passenger demand for a 10-minute interval and the corresponding number of screening staff required to achieve the 90-10 service criterion. The simulation was used intensively at this point. For each demand rate, a staffing level was selected and the simulation run at the constant staffing level and constant passenger demand rate. If the service criterion was met over the duration of the simulation, the staffing level was reduced; if not, it was increased. The simulation was run until the minimum staffing level to meet the service criterion was determined. This procedure was repeated until the look-up table was populated.
It became apparent that we needed to know how the staff were deployed between the wanding and bag-searching tasks, as well as how many equipment lines should be open and how the staff were distributed between them. Using the data in the throughput table, an "optimal staff allocation" defined as that with the highest maximum passenger throughput, was determined for each number of screening staff, at each PBS location (see Figure 3).
Figure 3: A graphical representation of an optimal staff allocation scenario. Different colors represent the number of lines that should be open for each total staffing level and the numbers within the bars represent the allocation of searchers and wanders to each security line.
Figure 4: Graphical representation of staff requirements of the optimal shift schedule for one scenario. The solid blue area represents the total staff requirements to achieve the target service level; the dashed line represents the total staff requirements plus surplus factor. The top line shows the staffing levels required by the optimal shift schedule.
Developing the Shift Schedules
The final task was to determine a minimum-cost shift schedule that achieves the required staffing levels. We aggregated the staffing levels across the four locations to obtain an airport-wide staffing requirement. Given a list of the possible shifts, we used a linear programming model to determine the optimal shift schedule that satisfied the airport-wide staffing requirements in each time period. A surplus factor was added to account for staff breaks. YVRAA management wanted the flexibility to evaluate a wide range of shift combinations including different durations and different start times. Our approach was sufficiently flexible to do this.
We believe that YVRAA was extremely far-sighted in recognizing the impact OR methods can achieve in improving security operations. This project gave students and faculty the opportunity to work on a challenging and significant applied project that set high standards for managing airport security operations.
The newly formed Canadian Air Transport Security Authority, the agency now responsible for passenger screening at 89 Canadian airports, is considering COE proposals to update this study for the current pre-board screening configurations and to extend the analysis to designing, operating and staffing the hold baggage screening operations.
For more on this and other applied projects, refer to www.coe.ubc.ca. Additional information about Vancouver International Airport is available at www.yvr.ca.
Mehmet Begen and Anita Parkinson received their M.Sc. degrees in Management Science through the COE program, and Bailey Kluczny received a B.Com degree from UBC Commerce. Begen is now a research analyst in the Centre for Operations Excellence, Parkinson is pursuing her Ph.D. at UBC, and Bailey is considering graduate studies. Derek Atkins and Martin L. Puterman are professors in the Operations and Logistics Division of UBC Commerce. Puterman is founder and director of the COE, and Atkins is associate vice president of planning for UBC.
OR/MS Today copyright © 2003 by the Institute for Operations Research and the Management Sciences. All rights reserved.
Lionheart Publishing, Inc.
506 Roswell Rd., Suite 220, Marietta, GA 30060 USA
Phone: 770-431-0867 | Fax: 770-432-6969
Web Site © Copyright 2003 by Lionheart Publishing, Inc. All rights reserved.