Cancellation Disruption Index Tool (CanDIT) Mona Kamal Mary Lee Brittlea Sheldon Thomas Van Dyke Bedis Yaacoubi Sponsor: Center for Air Transportation Systems Research (CATSR) Sponsor Contact: Dr. Lance Sherry George Mason University May 9, 2008 Overview • Problem • Background • Problem Statement • Solution • Data • Connectivity Factors • Passenger Factors • • • • Disruption Index Analysis Solver Conclusion Why this Project? • • • • • Problem Solution Data Connectivity Factors Passenger Factors • Disruption Index • Analysis • Solver • Conclusion Background Flight scheduling is a multi-step, water fall process Flight Schedule generation Fleet assignment Aircraft maintenance routing Crew Scheduling Yield Management OPERATIONS MANAGEMENT Background According to Bureau of Transportation Statistics (BTS) American Airline (14.8%)* % Cancelled SouthWest (12.2%)* % Cancelled United (11.5%)* % Cancelled Delta (10.8%)* % Cancelled 2003 2004 2005 2006 2007 2008 Average SD 1.61 1.78 1.45 1.57 2.83 2.70 1.99 0.61 1.01 1.02 0.85 0.81 0.85 0.80 0.89 0.10 1.09 1.18 1.30 2.05 2.43 2.62 1.78 0.67 1.05 1.56 2.69 1.52 1.37 1.49 1.61 0.56 * Market share based on revenue passenger miles for the year 2007 258 Domestic Flights Cancelled Per Day Average Stdev 1.57 % 0.65 % Possible Cancellation Scenarios • Flight cancellation due to mechanical problems • Cancellation initiated by the Airlines • Flight cancellation due to arrival restrictions, • Cancellation initiated by the Air Traffic Control • Flight cancellation due to safety restrictions, • Cancellation initiated by the FAA Scenario1:Flight cancellation due to mechanical problems Report a mechanical problem Provide feedback: Update is received Request the impact of canceling the flight Provide Disruption Factor of the flight Request impact of swapping flights Provide Disruption Factor for potential flights Provide prioritized cancellation strategy Provide appropriate decision PILOT/Maintenance Crew Airline Flight Cancellation Decision Tool Scenario 2:Flight cancellation due to arrival restriction Airport Arrival Demand saturation Request scheduled departing flights Show list of departing flights Request Disruption Indices for each departing flight to the low demand airport Provide Disruptions Indices for each flight Request prioritized flight cancellation decision Offer the prioritized flight disruptions Cancel low disruption flight AADC Airline Operations GUI Flight Cancellation Decision Tool Method for Cancellation • Currently, airline operations controllers rely on a Graphical User Interface (GUI) and Airport Arrival Demand Chart (AADC) to decide which flight to cancel. • Process is time consuming and may produce inefficient cancellation decisions. Operations Controllers GUI AADC Problem Statement Airlines schedule aircraft through multiple steps to connect passengers and crews. Flight cancellation scenarios may impact downstream flights and connections at a great expense. Given that cancellation is unavoidable, which flights should be cancelled to reduce airline schedule disruption and passengers inconvenience? Vision Statement A more sophisticated strategy for schedule recovery is needed to aid the controllers’ decisions and therefore avoid unnecessary costs to the airline. Once this system is implemented, controllers will have access to an automated decision support tool allowing them to reach low disruption cancellation decisions. Scope • Our focus is on two factors which lead to disruption : 1) The affect a canceled flight could have on other flights the same day 2) The reassignment of passengers on a canceled flight to other flights • We are considering disruption caused to ONLY the current day's schedule The Approach • • • • • Problem Solution Data Connectivity Factors Passenger Factors • Disruption Index • Analysis • Solver • Conclusion The team has … • • • • • Considered a single airline as the initial focus Looked at a one day flight schedule Determined connectedness of flights to one another Calculated a passenger reassignment factor Developed a disruption index which incorporates the effects of connectedness and passenger mobility • Created a tool, which uses these indices to determine the lower disruption flight(s) to cancel Disruption Index • End result • Decision making tool • A numerical value rating the disruption that the cancellation of a flight will cause to the airline for the remainder of the day • Combination of two factors: • Connectivity Factors • Passenger Factors Basis of our work • • • • • Problem Solution Data Connectivity Factors Passenger Factors • Disruption Index • Analysis • Solver • Conclusion Data • A spreadsheet was provided by the Study Sponsor containing the flight schedules of all domestic flights for one day • Information on all flights including: • • • • Carrier and tail number (i.e. airplane ID) Origin city and arrival city Scheduled departure and arrival times Actual departure and arrival times N444 Space Time Diagram SDF OAK LAS N781 MCI BNA N430 PHX BWI N730MA PIT SAN BDL N642WN HOU STL MDW PVD BHM OMA SLC 6:00 8:00 10:00 12:00 TIME 14:00 16:00 18:00 20:00 22:00 Statistics • Airline A • Fleet consists of more than 500 aircraft – Most are Boeing 737 aircraft • Each aircraft flies an average of 7 flights per day, totaling 13 flight hours per day • Serves 64 cities in 32 states, with more than 3,300 flights a day First Step: Connectivity • • • • • Problem Solution Data Connectivity Factors Passenger Factors • Disruption Index • Solver • Analysis and Conclusion Flight Connectivity • Definition: The transfer of passengers, crew, or aircraft from arriving at one destination to departing to the next within a designated time window IND SDF N444 2 hr connection window (8:30-10:30) N642WN More Flights No Flight N781 BNA BWI PVD MCI START END MDW N730MA BDL N430 BHM SAN ISP 6:00 7:00 8:00 9:00 TIME 10:00 11:00 12:00 Connectivity Factors (CFs) • Connectivity factors determines the number of down-path flights that could be impacted by the cancellation of a single flight • Each flight leg is assigned a connectivity factor 100% Flight Connectivity • Arriving flights connect to all flights that are scheduled to depart from that airport within a designated connection window. Assumptions: [1]: There is at least one passenger or crew member on an arriving flight that will have to board a departing flight. [2]: Connecting flights must be assigned a minimal time for passengers to physically transfer from the arriving flights. Flight Connectivity (CF) Factors N444 BWI 7 4 N781 1 5 N642WN PHX 3 1 3 3 1 7 IND 2 SAT 1 N730MA Flight Connectivity (CF) Factors N444 BWI 7 4 N781 1 5 N642WN PHX 3 1 3 3 1 7 IND 2 SAT 1 N730MA Flight Connectivity (CF) Factors N444 BWI 7 4 1 5 N781 1 N642WN PHX 3 1 3 3 1 7 IND 2 SAT 1 N730MA Flight Connectivity (CF) Factors N444 BWI 7 4 1 5 N781 1 N642WN PHX 3 1 3 3 1 7 IND 2 2 SAT 1 N730MA Flight Connectivity (CF) Factors N444 BWI 7 4 1 5 N781 1 N642WN PHX 3 1 3 3 3 1 7 IND 2 2 SAT 1 N730MA Flight Connectivity (CF) Factors N444 BWI 7 4 1 5 N781 1 N642WN PHX 3 4 1 3 3 3 1 7 IND 2 2 SAT 1 N730MA Flight Connectivity (CF) Factors N444 BWI 7 4 1 5 5 N781 1 N642WN PHX 3 4 1 3 3 3 1 7 IND 2 2 SAT 1 N730MA Flight Connectivity (CF) Factors N444 BWI 7 4 6 1 5 5 N781 1 N642WN PHX 3 4 1 3 3 3 1 7 IND 2 2 SAT 1 N730MA Flight Connectivity (CF) Factors N444 BWI 7 7 4 6 1 5 5 N781 1 N642WN PHX 3 4 1 3 3 3 1 7 IND 2 2 SAT 1 N730MA 100% flight connectivity [45min,120min] Top 3 flights are connected to 55% of the flights throughout the day. All 3 flights leave close to 6:30 and are headed to MDW Total flights during this day is 1853 1100 1000 Connectivity Factor 900 800 700 A Flight arriving at small airport, ORF at 8:40 has low connectivity 600 500 400 300 Flights destined for airports with less traffic have low connectivity 200 100 0 0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 Scheduled Arrival Time 0:00 3:00 6:00 100% connectivity: Sensitivity Analysis The connection window was varied over 5 more time intervals: [45* min, 120 min] [45 min, 150 min] [45 min, 180 min] (Baseline) [45 min, 210 min] [45 min, 240 min] *The minimal time window was fixed at 45 minutes for this study, as a reasonable amount of time for physical transfer of passengers Varying Connection windows Varying Connection windows Connection window: 240 min max vs. 120 min max 180 min max vs. 150 min max 1200 1200 1000 45 min to 240 min window y = 1.0453x R² = 0.9966 45 min to 180 min window 800 600 400 200 800 y = 1.1795x R2 = 0.9772 600 400 200 0 0 0 200 400 600 800 1000 1200 1200 y = 1.0224x R2 = 0.998 1000 800 600 400 200 0 0 200 400 600 800 45 min to 180 min window 1000 0 200 400 600 800 45 min to 120 min window min180 to 150min min window 210 min max45vs. max 45 min to 210 min window 1000 1200 1000 1200 Partial Connectivity • Realistically, flights are connected at different rates based on the airline strategy (hub and spoke or focus cities …), the connecting airport , and other factors. • A study led by Darryl Jenkins on Airline A developed % passengers connectedness at all airports. • The data used in the study: Average Outbound, non interline passengers (Pax) from each city (from O & D Database) Average enplaned Pax from each city (from the Onboard Database) Airport Percent Connect Year of 2002 Data Author divides airports to : 1. Major connecting airports 2. Partial Connecting airports 3. Non-connecting airports http://www.erau.edu/research/BA590/chapters/ch1.htm Airports % connect HOU 29.0% MDW .…. 23.5% ….. .…. ….. JAX 12.4% AUS .…. 10.7% ….. .…. ….. ALB 0.4% BDL 0.0% Flight Connectedness We then incorporated the Airport Percent Connect (APC) data to our CF generator algorithm: if APC >= 15 % , then 100% connect if APC < 2%, then 0 % Connect if 2%<APC<15%, then [(APC- 2) * 100 / 13 ] % Connect Comparing Graphs from the two methods Low CF for early flight 100 % Flight Connectivity APC Flight Connectivity Comparing APC and 100% Connectivity Comparing results from the two methods Tail number Leg Num origin1 dest1 Scheduled Schedule in cf_45_180 cf_45_180 out time time 100% APC N683 2 RNO LAS 8:00 9:10 527 462 N632 2 RNO PDX 8:05 9:25 292 118 N617 2 RNO SEA 8:30 10:15 250 127 N687 3 RNO LAX 9:10 10:35 378 228 N649 1 RNO SLC 10:05 12:25 238 182 N651 3 RNO LAS 10:15 11:25 312 280 Table 2: Least disruptive (considering only connectedness) flight based on 100% Connectivity and Airport Percent Connect Algorithm on other airlines Airline B 1200 1100 1100 1000 1000 900 900 Connectivity Factor 1200 800 700 600 500 400 800 700 600 500 400 300 300 200 200 100 100 0 0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00 3:00 0 0:00 6:00 Airline C 6:00 9:00 12:00 15:00 18:00 21:00 0:00 Three different airlines with 100% connectivity within a 45 to 180 minute time window 1200 1100 1000 900 800 700 600 500 400 300 200 100 0 0:00 3:00 Arrival Time Arrival Time Connectivity Factor Connectivity Factor Airline A 3:00 6:00 9:00 12:00 15:00 Arrival Time 18:00 21:00 0:00 Second Factor • • • • • Problem Solution Data Connectivity Factors Passenger Factors • Disruption Index • Analysis • Solver • Conclusion Passenger Factor • Takes into consideration number of passengers on flight as well as remaining seats that day • Equation: Number of Passengers on Flight Total Number of Available Seats • Higher penalty for a higher ratio Passenger Factor • No data available on number of passengers and capacity of individual flights • Formula fully functional so airline can input flight information • For analysis purposes, used a random number generator Putting It All Together • • • • • Problem Solution Data Connectivity Factors Passenger Factors • Disruption Index • Analysis • Solver • Conclusion Calculation of Disruption Index • Disruption Index • = W1(ConnFact) + W2 (α)(PaxFact) W1 and W2 = Weights given to each factor (a one time setting for each airline) α = Scaling factor for passengers Spreadsheet Solver How it All Works • • • • • Problem Solution Data Connectivity Factors Passenger Factors • Disruption Index • Analysis • Solver • Conclusion Functionality Test • Algorithm tested for functionality using historical data • Different airlines tested, each with different schedule date • Shows how airline would use this data PF Tail # Departure Origin Destination Time Arrival Time DI Weighted CF Weigh ted PF N343NB MSP SLC 9:11 11:04 1 0.5 2.4 0.4 N301US MSP MCO 10:21 14:19 1 0.5 2.6 0.5 N313US MSP SMF 9:16 11:07 1 0.5 2.8 0.5 N596NW MSP PDX 9:30 11:20 2 1.5 1.5 0.3 N362NB MSP IAH 9:10 11:53 2 1.5 1.9 0.3 N348NB MSP EWR 10:47 14:17 2 1.5 2.1 0.4 N327NW MSP SJC 10:20 12:23 2 0.5 7.9 1.4 N375NC • • MSP • • RSW • • 10:18 • • 14:28 • • 2 • • 1.5 • • 3.6 • • 0.6 • • N378NW MSP TPA 10:23 14:12 6 5.0 4.1 0.7 N777NC MSP MEM 10:14 12:07 6 6.0 1.0 0.2 N8925E MSP MKE 10:08 11:08 6 6.0 1.2 0.2 N780NC MSP DTW 10:06 12:52 8 7.5 0.5 0.1 24.5 21.1 N338NW MSP PSP 9:20 11:01 26 2.0 135. 1 N303US MSP MIA 10:30 14:48 35 14.0 116.5 Destinati Departure on Time Arrival Time DI Weighted CF PF Weighted PF Tail # Origin N171US CLT SFO 9:16 12:03 12 1.5 3.1 10.8 N514AU CLT ORF 9:26 10:29 13 3.5 2.8 9.7 N449US CLT BUF 9:19 10:47 17 12.0 1.5 5.2 N525AU N439US CLT CLT SRQ MIA 9:24 9:54 11:12 12:01 23 24 2.0 20.5 6.0 1.1 20.5 3.7 N749US N453UW N426US CLT CLT CLT DEN BWI JAX 9:38 8:03 9:45 11:31 9:22 10:59 27 29 32 23.5 26.5 28.5 1.1 0.8 1.0 3.8 2.8 3.4 N530AU CLT DFW 9:28 11:20 35 28.5 1.9 6.4 N459UW • • N918UW CLT • • CLT PBI • • LAS 9:37 • • 9:47 11:43 • • 11:24 35 • • 71 28.5 • • 66.5 1.9 • • 1.4 6.6 • • 4.9 N533AU CLT DFW 8:09 9:55 72 66.5 1.5 5.1 N939UW N922UW CLT CLT MCO TPA 9:58 9:52 11:36 11:37 79 79 77.0 75.0 0.6 1.2 2.1 4.2 N457UW CLT PHL 9:30 10:58 82 79.0 0.8 2.9 N721UW CLT BOS 8:08 10:14 94 91.0 0.8 2.9 N574US CLT CMH 9:59 11:16 189 2.0 54.2 187.2 Solving Tool • • • • • Problem Solution Data Connectivity Factors Passenger Factors • Disruption Index • Analysis • Solver • Conclusion Tom’s Solver hyperlink Solving Tool • • • • • Problem Solution Data Connectivity Factors Passenger Factors • Disruption Index • Analysis • Solver • Conclusions Conclusions • Created an index that assigns a numerical value based on the degree of disruption in the system • Developed a tool to allow controllers to make better informed decisions • Tool can be easily modified to incorporate factors not previously considered • Tool will allow users to make an educated decision based on the disruption of a flight • Reduces time to make decision and may improve customer satisfaction Future Works • Consider crew connectivity • Consider other factors in disruption index not previously considered (such as cost) • Consider flight interconnectivity • Consider linking tool to web to attain real time data • Considering more than just a single day schedule References • • • • http://www.isr.umd.edu/airworkshop/ppt_files/Ater.pdf Images: http://fly.faa.gov/Products/AADC/aadc.html http://ocw.mit.edu/NR/rdonlyres/Civil-and-Environmental-Engineering/1206JAirline-Schedule-PlanningSpring2003/582393E6-2CA6-4CC1-AE661DAF34A723EA/0/lec11_aop1.pdf Embry-Riddle Aeronautical University • http://www.erau.edu/research/BA590/chapters/ch1.htm Question Backup-Varying Connection windows Connection Window: 45 to 180 Minutes Connection window: 45 to 180min 1100 1200 1000 1100 900 1000 800 900 Connectivity Factor Connectivity Factor Connection Window: 45 to 150 Minutes Connection window: 45 to 150min 700 600 500 400 300 200 100 0 0:00 800 700 600 500 400 300 200 100 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00 3:00 0 0:00 6:00 3:00 6:00 9:00 Time 3:00 6:00 Connection Window: 45 to 240 Minutes Connection window: 45 to 240min 1200 1200 1100 1100 1000 1000 900 Connectivity Factor 900 Connectivity Factor 0:00 Time Connection Window: 45 to 210 Minutes Connection window: 45 to 210min 800 700 600 500 400 300 200 800 700 600 500 400 300 200 100 100 0 0:00 12:00 15:00 18:00 21:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 Time 0:00 3:00 6:00 0 0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 Time 0:00 3:00 6:00 Investigating Connectedness-Sensitivity The highest 10 increases in CF by percent based upon adding 30 minutes to the connection window: 1. 2. 3. Origin Destination Departure Arrival Destination Size1 BWI BUF 09:55 11:00 PHX ELP 08:15 PHX ELP MDW CF12 CF23 23 1 60 10:35 19 1 78 10:55 13:00 19 1 80 DTW 10:40 12:45 25 1 95 MDW OMA 09:45 11:05 28 1 117 TPA MSY 08:50 09:25 34 1 157 BWI RDU 07:15 08:20 38 1 175 BNA CLE 07:30 09:55 36 1 176 MDW IND 06:45 07:40 24 1 185 TPA JAX 07:15 08:05 17 1 251 In this case size refers to the total number of entering and departing flights from the airport CF1 is the connectivity factor for a 45 to 150 minute connection window. CF2 is the connectivity factor for a 45 to 180 minute connection window Airport Percent Connect CFs Connection Window: 45 to 180 Minutes Accounting for Passenger Connections 1100 1000 Low CF for early flight Connectivity Factor 900 800 700 600 500 400 300 200 100 0 0:00 3:00 6:00 9:00 12:00 15:00 Time 18:00 21:00 0:00 3:00 6:00 • EVM • WBS • GANNT Window chosen for analysis • For analysis purposes, chose • [45 min, 180 min] • The airline may choose a connectivity window which fits their flight patterns best • The time window is an appropriate cut-off because the values … Generalizing Algorithm • Data for two more airlines has been compiled • Connectivity factors have been computed • Airports differ for each airline • Partial-connection percentages have only been found for the first airline (Airline A) • Known airports have been assigned same connection percentage as from the first airline • Unknown airports have been given a default connection percentage Percent Connectivity Airline B Connectivity Factors, 100% Connectivity 1000 900 900 800 800 Connectivity Factor Connectivity Factor 1000 700 600 500 400 300 700 600 500 400 300 200 200 100 100 0 0:00 3:00 6:00 9:00 12:00 15:00 Arrival Time 18:00 21:00 0:00 Connectivity Factors, Percent Passenger Connectivity 0 0:00 3:00 6:00 9:00 12:00 15:00 18:00 21:00 0:00 Arrival Time As before, accounting for percent connectivity had a significant effect on the outputs. A similar decrease in data occurred for Airline C Agents/Stakeholders • Airline Operations Control • FAA • Air traffic controllers • Passengers • Pilots/flight crew • Maintenance crew