Toeval is logisch. Johan Cruijff

Size: px
Start display at page:

Download "Toeval is logisch. Johan Cruijff"

Transcription

1 Toeval is logisch Johan Cruijff

2 Platform Bus Simulation Barbara Pieters, Marjan van den Akker Department of Information & Computing Sciences

3 Outline Problem description Simulation model Objective Model description Validity, Credibility & Verification Input & OutputT Simulation study Experiments Results & Conclusions

4 Gate- and busplanning at AAS At Amsterdam Airport Schiphol two types of gates: With airbridge to terminal Platform stands: passengers are transported by Platform bus

5 Platform Bus Planning

6 Platform Bus Planning Set of flights assigned to a platform stand: Arrival and departure time Number of passengers Platform stand location/number Entree and exit point at terminal building Corresponding set of required bus trips: One flight leads to several bus trip Arrival: all busses at the same time Departure: sequence of at least two bus trips Collection of bus-driver shifts There are two types of buses: Cobus: 70 passengers City bus: 50 passengers Some trips must be performed by a cobus Bus drivers need a break during their shift (labour legislations)

7 Platform Bus Planning: BAAS For upcoming day bus trips have to be assigned to shifts Goal is to maximize robustness Ability to cope with disturbances during the day of operation Cost function for non-robustness: Depends on idle times between two subsequent trips High for small idle times Low for long idle times Descending steeply in beginning

8 Why Simulate? (Simple) Testcase Low cost Compare schedulers BAAS and BusCom BusCom: used at Amsterdam Airport Schiphol BAAS: developed by Guido Diepen at UU

9 BAAS vs BusCom BusCom First Come First Served Frequent rescheduling Frequent system crashes Frequent human intervention needed

10 BAAS vs BusCom BusCom First Come First Served Frequent rescheduling Frequent system crashes Frequent human intervention needed Busplanner Amsterdam Airport Schiphol (BAAS) Column generation method Less frequent rescheduling? More robust schedules?

11 Elements In the Model Flights (type, time, passengers, platform,...) Bus Trips ((bus)type, time, platforms,...) Bus/driver shifts (bustype, times, amount,...) Bus plans: sequence of trips performed by one bus (bustype, times, trips,...)

12 A Schedule... FCFS BAAS

13 States & Events - first idea

14 States & Events - first idea States Bus is busy or idle Trip is being driven/completed/available/unavailable... Events Flight arrives Scheduled departure of passenger from terminal for departing flight. Trip starts Trip finishes

15 However, our concern is the planning

16 However, our concern is the planning State Busplan (schedule) Initial Trip A Trip B Trip C Infeasible Trip A Trip B Trip C Trip A Trip C Trip B Assumption:

17 However, our concern is the planning State Busplan (schedule) Initial Trip A Trip B Trip C Infeasible Trip A Trip B Trip C Trip A Trip C Trip B Assumption: for given flight arrival and departure times, busses drives according to the plan.

18 Events

19 Events Flight time update (Central Information System Schiphol, we assume a flight is updated once) 1. Update busplan 2. If infeasible busplan

20 Events Flight time update (Central Information System Schiphol, we assume a flight is updated once) 1. Update busplan 2. If infeasible busplan Reschedule!; If schedule contains unassigned trip schedule Unassigned trip is due Unassigned trip is due If trip has not been assigned in the mean time, Try emergency measures. Emergency measures

21 Events Flight time update (Central Information System Schiphol, we assume a flight is updated once) 1. Update busplan 2. If infeasible busplan Reschedule!; If schedule contains unassigned trip schedule Unassigned trip is due Unassigned trip is due If trip has not been assigned in the mean time, Try emergency measures. Emergency measures Delete trip Change break time Change shift end time Fix trip to busplan

22 Events Flight time update (Central Information System Schiphol, we assume a flight is updated once) 1. Update busplan 2. If infeasible busplan Reschedule!; If schedule contains unassigned trip schedule Unassigned trip is due Unassigned trip is due If trip has not been assigned in the mean time, Try emergency measures. Emergency measures Delete trip Change break time Change shift end time Fix trip to busplan If unable to assign trip: remove from simulation!

23 Validity, Credibility & Verification

24 Validity, Credibility & Verification Interviews with experts (Schiphol, G. Diepen) Extensive code review and debugging Testruns Implement graphical component Define experiments and output

25 Input

26 Input Arrival/Departure times (and disruptions) Arrival/Departure information (flights, passengers,...) Driving time information (distances) Bus information (amount, shift information,...)

27 Arrival Time Disruptions

28 Arrival Time Disruptions Normal distribution: µ = 4 minutes and σ 2 = 30 minutes.

29 Departure Time Disruptions

30 Departure Time Disruptions Exponential distribution. Delayed departures: 0.98 exp(15) Early departures: 0.02 exp(2)

31 Output Objective: Compare BusCom to BAAS. Hypothesis: BAAS schedules are more robust, less rescheduling is needed, fewer conflicts arise.

32 Output Objective: Compare BusCom to BAAS. Hypothesis: BAAS schedules are more robust, less rescheduling is needed, fewer conflicts arise. Monitor: # rescheduling Types and # emergency measures Robustness of schedule

33 Robustness - Cost Function idle time(trip a, trip b )= T start trip b T end trip a driving time(trip a, trip b ), c B (idle) = 1000(arctan( 0.21 idle) + π 2 n m ), Robustness of schedule: c B (idle), i=0 j=0 n = # bus plans; m = # trip pairs

34 Designing Experiments Use real schedules, choose subset (5 days) Use approximation distributions for disruptions Record important states and events # Rescheduling # and types of emergency measures Robustness of schedules (start, intermediate, end)

35 Designing Experiments Use real schedules, choose subset (5 days) Use approximation distributions for disruptions Record important states and events # Rescheduling # and types of emergency measures Robustness of schedules (start, intermediate, end) Use same settings for both schedulers 25 runs per day

36 Results - 1 Used statistics: mean, standard deviation, ratio,... Robustness day 5 day 15 day 20 day 25 day 30 µ BAAS FCFS BAAS FCFS

37 Results - 1 Used statistics: mean, standard deviation, ratio,... Robustness day 5 day 15 day 20 day 25 day 30 µ BAAS FCFS BAAS FCFS # Rescheduling day 5 day 15 day 20 day 25 day 30 µ BAAS FCFS BAAS FCFS

38 Results - 2 # Measures day 5 day 15 day 20 day 25 day 30 µ BAAS FCFS BAAS FCFS

39 Results - 2 # Measures day 5 day 15 day 20 day 25 day 30 µ BAAS FCFS BAAS FCFS

40 Conclusions BAAS performs better Less rescheduling More robust schedules Less emergency measures used

41 Conclusions BAAS performs better Less rescheduling More robust schedules Less emergency measures used Lessons learned (according to Barbara Pieters) Define simulation objective carefully Construct simulation model carefully Use simulation package/library if possible/feasible If not: reserve time for debugging Make simulation graphical if possible