MIT MIT ICAT ICAT MIT International Center for Air Transportation Virtual Hubs: A Case Study Michelle Karow karow@mit.edu John-Paul Clarke johnpaul@mit.edu MIT ICAT Presentation Overview: • Motivation • Definition • Characteristics • Problem formulation • Application at a major US carrier • Limitations and future considerations MIT ICAT Irregular operations at a hub airport can be crippling to an airline schedule • Reduction in capacity typically necessitates cancellations and delays • Effects resonate network-wide and on all levels of operation (fleet, maintenance, crew and passengers) • Majority of irregularities caused by weather Could airlines reduce the number of delays and cancellations by rerouting entire connecting banks to an airport with excess capacity? MIT ICAT Re-directing flights through a virtual hub can provide relief to the original hub with minimal disruption • Shift connecting demand over two hubs, decreasing strain on the original hub • Continuity of passenger flow, insuring a reduction in total passenger delay • Capitalize on under-utilized airports Definition: A virtual hub is a predetermined alternative airport that during irregular operations at the original hub, hosts connection complexes to maximize passenger flow through the network. MIT ICAT Sample virtual hub network Origin Original Hub Passengers destined for the hub Origin Origin Passengers connecting to destinations not served by the virtual hub Destination Destination Destination Destination Origin Destination Origin Origin Origin Destination Virtual Hub Destination MIT ICAT Virtual hubs can be identified by the following characteristics: • Low average daily delays Check FAA’s Airport Capacity Benchmark report for delay rankings of US airports • Geographically equivalent location to the original hub Geographical location Average Delays Virtual Hub Candidates Check relative location to existing hub •Excess capacity Track airline gate utilization throughout the day, given low delays indicate excess airport capacity Excess Capacity Virtual Hub MIT ICAT Implementing a virtual hub network consists of two phases: The Virtual Hub Model and The PRM Virtual Hub Model Accommodated Passengers Disrupted Passengers Add to the next time window Passenger Re-accommodation Module (PRM) Passengers that can be reaccommodated (and itineraries) Passengers that cannot be accommodated MIT ICAT Phase I: Implementing a virtual hub network Anticipated Weather/ Ground Delay Program •Implemented in the hours before the weather is predicted to impact the operations at the original hub • Solved iteratively over connecting bank timewindows until weather has cleared Update Variables for Next Time Window • Maximizes passenger flow, in turn minimizing total passenger delay Airport Capacities Passenger Itineraries Aircraft Capacities Time Window t1 Time Window t2 Maximize Passenger Flow Maximize Passenger Flow Original Hub Flights Virtual Hub Flights Original Flight Schedule Time Window tn …. Delayed/ Cancelled Flights Maximize Passenger Flow Adjusted Itineraries MIT ICAT Key Assumptions: • Ground resource availability • Crew and maintenance flexibility • Passenger connections within a time window • Passenger consent MIT ICAT The virtual hub model is formulated as a mixed integer network flow problem. Input data: • Size of the time windows • Passenger itineraries • Original flight schedules • Airport capacities • Aircraft capacities MIT ICAT Objective function: Maximize passenger flow Maximize dij zijk iO jD kH Where: O D H dij zijk set of origins set of destinations set of hub airports {OH, VH, VHs} demand from origin i to destination j positive variable representing the fraction of demand traveling on the network from origin i to destination j through hub k MIT ICAT Subject to: Definition of zijk: A path exists from origin to destination through a hub zijk wijk i O, j D,k H wijk xik wijk yik i O, j D, k H wijk xik ykj 1 z kH ijk 1 i O, j D, k H i O, j D, k H i O, j D Where: wijk binary decision variable that the network exists from origin i to destination j through hub k xik binary decision variable that the network exists from origin i to hub k ykj binary decision variable that the network exists from hub k to destination j MIT ICAT Subject to: Airport capacity: Upper bounds on aircraft sent to a hub x iO ik ck k H Where: xik binary decision variable that the network exists from origin i to hub ck capacity of hub k MIT ICAT Subject to: Aircraft Capacity: Upper bounds on the number of passengers on an aircraft dij zijk pi i O dij zijk q j j D jD kOH ,VH iO kOH ,VH d jD d iO z fi i O, k VH s z gj j D, k VH s ij ijk ij ijk Where: dij demand from origin i to destination j zijk binary decision variable that the network exists from origin i to destination j through hub k pi, qj aircraft capacity to and from the hub, respectively fi,, gj excess aircraft capacity on scheduled flights to and from the virtual hub, respectively MIT ICAT Subject to: Hub choice: A flight is served either by the virtual hub or the original hub kOH ,VH ykj 1 j D kOH ,VH xik 1 i O Conservation of Flow: Upper bounds on aircraft departures from hubs x y iO ik jD kj bk 0 k Where: xik ykj bk binary decision variable that the network exists from origin i to hub k binary decision variable that the network exists from hub k to destination j number of aircraft on the ground from the previous time window at hub k MIT ICAT Phase II: Re-accommodating disrupted passengers After the scheduling decisions are made for a time window, some passengers will be disrupted and require re-accommodation. Disrupted passengers for the virtual hub network include the following: •A connecting passenger with their original flight from their origin serviced by the virtual hub and their original flight to their destination serviced by the original hub. •A connecting passenger with their original flight from their origin serviced by the original hub and their original flight to their destination serviced by the virtual hub. •A non-stop passenger with their original flight either to or from the original hub serviced by the virtual hub. MIT ICAT 1st Leg diverted to VH + 2nd leg on VHs Destined for OH Accommodated on a later flight to OH Originating at OH Accommodated on a later flight from OH 1-leg itinerary Disrupted Passengers from Virtual Hub Model 1st Leg diverted to VH 2-leg itinerary 1st Leg on VHs + 2nd leg rescheduled from VH 2nd Leg rescheduled from VH Accommodated on a later flight from VH Accommodated on a later flights through OH Accommodated on a later flights through OH Accommodated on a later flight to VH Re-accommodated Passengers An overview of the Passenger Re-accommodation Module (PRM) MIT ICAT A closer look: Application of the Virtual Hub Network to a Major US Carrier A thunderstorm was present at the original hub airport on March 9, 2002 while the virtual hub remained relatively unaffected. For this day, throughout the network: Domestic and International Flights 4,000 Number of Passengers 99,000 Distinct Itineraries 38,000 MIT ICAT Major delays plague the original hub while relatively minor effects are felt at the virtual hub Delayed Flights per Hub on March 9, 2002 180 Flights delayed >15 minutes 160 Number of Flights 140 Flights delayed >30 minutes 120 Flights delayed >45 minutes 100 80 Flights delayed >60 minutes 60 40 Cancelled flights 20 0 OH Departures OH Arrivals VH Departures VH Arrivals MIT ICAT Input data: Size of the Time Window The two-hour time window was selected to accommodate both the need for high scheduling accuracy and a large percentage of passengers connecting in distinct time windows. Average Connection Time 151 minutes Highest Frequency Markets 1 flight per 60 minutes Size of the Time Window 120 minutes MIT ICAT Input data: Passenger Itineraries • Only the flight legs originating or arriving at the original hub were considered. • Itineraries with international flight legs were treated as originating or arriving at the original hub • Itineraries with connections overlapping two time windows were separated into two itineraries, originating and arriving at the original hub Traveling through the original hub during the period of irregular operations Itineraries Passengers 4,342 19,291 MIT ICAT Input data: Original Flight Schedules • Only domestic flights are eligible for diversion to the virtual hub • International flights operated by the airline are assumed to depart or arrive within one time window of their schedule. • International flights operated by the airline’s code-share partners are also assumed to depart or arrive within one time window of their schedule. Flights between 8am and 6pm at the original hub Domestic International 548 46 MIT ICAT Input data: Virtual Hub Airport Capacities • Track cumulative operations at the virtual hub airport throughout the day • Bias the data to produce positive aircraft totals at the airport throughout the day (account for aircraft kept overnight) • Subtract the number of operations at the airport from the number of gates to find the excess capacity per time window MIT ICAT Throughout the day, the virtual hub is does not reach it’s maximum gate capacity of 45 gates Cumulative Number of Aircraft for the Airline at the VH on March 9, 2002 45 40 30 25 20 15 10 5 Time in Hours 00 24 00 23 00 22 00 21 00 20 00 19 00 18 00 17 00 16 00 15 00 14 00 13 00 12 00 11 0 00 10 90 0 80 0 70 0 60 0 50 0 0 40 Number of Aircraft 35 MIT ICAT Subtracting the cumulative number of aircraft from the total number of gates provides a measure of excess capacity Excess Capacity for the Airline at the VH on March 9, 2002 50 45 35 30 25 20 15 10 5 Time in Windows 00 24 00 23 00 22 00 21 00 20 00 19 00 18 00 17 00 16 00 15 00 14 00 13 00 12 00 11 0 00 10 90 0 80 0 70 0 60 0 50 0 0 40 Number of Aircraft 40 MIT ICAT The excess capacity over the day is compressed into two hour time windows to determine the VH excess capacity during irregular ops 100 Excess Capacity for the Airline at the VH on March 9, 2002 90 Number of Aircraft 80 70 60 50 40 30 20 10 0 800 -1000 1001-1200 1201-1400 Time Windows 1401-1600 1601-1800 MIT ICAT Input data: Virtual and Original Hub Airport Capacities • The capacity at the original hub was reduced by 1/3 to reflect the reduction in the airport arrival rate required by the ground delay program. • The capacity at the virtual hub was the minimum number of gates to accommodate all diverted flights. Time Window Scheduled Domestic Arrivals Scheduled Domestic Departures cOH: Original Hub Capacity cvh: Virtual Hub Capacity 800 to 1000 35 57 21 19 1001 to 1200 41 58 28 19 1201 to 1400 47 42 32 19 1401 to 1600 59 53 40 19 1601 to 1800 33 37 22 19 MIT ICAT Input data: Aircraft Capacities • Flights remain assigned to their originally schedule aircraft, regardless of which hub airport they are sent to. • Capacity for flights traveling through the original hub is the number of seats on the aircraft. • Capacity for scheduled flights through the virtual hub is the number of seats minus the number of passengers booked on the flight (i.e., excess capacity). MIT ICAT Phase I Implementation: The Virtual Hub Model • Solution times for the time windows range from 5 minutes to over an hour, depending on the sparsity of the data set. • In each time window, the maximum number of aircraft were sent to the original hub. Time Window Number of Passengers Constraints Variables Passengers Served (Objective Function) 800 to 1000 4,436 26,304 12,247 4,037 1001 to 1200 6,191 31,311 14,566 5,747 1201 to 1400 5,139 26,019 12,112 4,753 1401 to 1600 6,298 41,100 19,099 5,852 1601 to 1800 3,122 16,639 7,762 2,978 MIT ICAT Phase II Implementation: PRM • Passengers (and itineraries) not accommodated by the virtual hub model were entered into the PRM after each time window. • International passengers were considered disrupted if their domestic leg was delayed by more than 4 hours (i.e., two time windows). • Un-accommodated passengers are passengers that could not be accommodated by the end of the day on flights traveling through either hub airport. Time Window Passengers Not Accommodated by Virtual Hub Model Re-accommodated Passengers Disrupted International Passengers Un-accommodated Passengers 800 to 1000 399 340 53 6 1001 to 1200 444 321 107 16 1201 to 1400 386 361 21 4 1401 to 1600 446 356 58 32 1601 to 1800 144 131 9 4 MIT ICAT Comparing Actual Recovery to the Virtual Hub Network Actual Recovery Virtual Hub Network Total Passengers 19,291 19,291 Number of Cancelled Flights 123 0 Passengers Requiring ReAccommodation 774 1,665 Disrupted International Passengers 237 248 Un-Accommodated Domestic Passengers 207 67 Passengers Delayed Over Two Hours 14,123 838 94% reduction MIT ICAT Limitations and Future Considerations: • Number of airline gates is somewhat flexible; cannot ensure airports will maintain good virtual hub candidacy. •Crew constraints and contract conditions could limit feasibility and increase diversion costs. • Availability of ground resources may constrain the capacity of the virtual hub. • Iterating over time windows under-estimates abilities of weather forecasting while optimizing over multiple time windows adds complexity and non-linearity. •Consideration of re-accommodating passengers on scheduled non-stop flights will provide a better (or equivalent) solution.