1. CREW SCHEDULING Past and Future Jacques Desrosiers HEC & GERAD Montréal, Canada

2. Voir la présentation Power Point Crew Sched - Complements.ppt pour deux autres transparents

3. The Mathematics behind Vehicle Routing and Crew Scheduling Jacques Desrosiers HEC & GERAD Montréal, Canada Canadian Mathematical Society Montréal, December 11 -13, 1999

5. OUTLINE  Problem Structure  Solution Approaches  Successful Applications  Research Trends  Conclusions

6.  PROBLEM STRUCTURE Time-Space Networks Local Schedule Restrictions Task Covering Schedule Composition Non Linear Cost Functions

7. Time-Space Networks

8.  SOLUTION APPROACHES Branch & Cut involving... Lagrangean Relaxation Dantzig-Wolfe Decomposition Kelley’s Cutting Plane Method Column Generation

9. Resource Constrained Shortest Path Problem on G=(N,A) P(N, A) :

10. Integer Multi-Commodity Network Flow Structure

11. Crew Scheduling Problems Sub-Problem is Strongly NP- hard Does not possess the Integrality Property Master Problem : Set Partitioning/Covering

12. Branching & Cutting Decisions Branch-and-Cut Decomposition Process applied at all decision nodes

13.  SUCCESSFUL APPLICATIONS Vehicle Routing with Time Windows Dial-a-Ride for Physically Disabled Persons Urban Transit Crew Scheduling Multiple Depot Vehicle Scheduling Aircraft Routing Crew Pairing Crew Rostering (Pilots & Flight Attendants) Locomotive and Car Assignment

14. CREW-OPT BUS-OPT ALTITUDE-Pairings ALTITUDE-Rosters ALTITUDE-PBS RAIL-WAYS The GENCOL Optimizer 60 installations around the world … at the Core of Various Software Systems

15.  RESEARCH TRENDS Accelerating Techniques Primal - Dual Stabilization Constraint Aggregation Sub-Problem Speed-up Two-level Problems Solved with Benders Decomposition Integer Column Generation with Interior Point Algorithm

16. Acceleration Techniques Column Generator Master Problem Global Formulation Heuristics Re-Optimizers Pre-Processors …to obtain Primal & Dual Solutions

17. Acceleration Techniques... Multiple Columns: selected subset close to expected optimal solution Partial Pricing in case of many Sub-Problems Early & Multiple Branching & Cutting: quickly gets local optima Branching & Cutting: on integer variables !

18. Primal - Dual Stabilization Restricted Dual Perturbed Primal Stabilized Primal

19. Dual SolutionPrimal Solution Primal SolutionDual Solution Approximate Primal & Dual Primal & Dual Solutions Primal - Dual Stabilization...

20. Constraint Aggregation Massive Degeneracy on Set Partitioning Problems A pilot covers consecutive flights on the same aircraft A driver covers consecutive legs on the same bus line Aggregate Identical Constraints on Non-zero Variables

21. Aggregation Algorithm Initial Constraint Aggregation Consider only Compatible Variables Solve Aggregated Master Problem Primal & Aggregated Dual Solutions Dual Variables Split-up Solve Sub-Problem Modify Constraint Aggregation

22. Sub-Problem Speed-up Resource Constrained Shortest Path Labels at each node : cost, time, load, … Resource Projection Adjust A dynamically Generalized Lagrangian Relaxation Results on Sub-Problem cpu time divided by 5 to 10

23. Two-Level Problems Benders Decomposition Algorithm for Simultaneous Assignment of Buses and Drivers Aircraft and Pilots Pairings and Rosters Locomotives and Cars

24. IP(X, Y) for Two-Level Scheduling MIP(X, y) solved using Benders Decomposition Master IP(X) Simplex and B&B(X) Sub-Problem solved by Column Generation MP LP(y) of Set Partitioning SP DP for Constrained Paths B&B(Y) with MIP(X, y) at each node

25. Benders MP Benders SP B & B IP LP DP CG MP CG SP

26. Column Generation with Interior Point Algorithm ACCPM Algorithm (Goffin & Vial) Applications Linear Programming Non-Linear Programming Stochastic Programming Variational Inequalities

27. Integer Column Generation with Interior Point Algorithm Strategic Grant in Geneva –J.-P. Vial et al. Strategic Grant in Montréal –J.-L. Goffin et al. Design of a Commercial Software System

28.  CONCLUSIONS Larger Problems to Solve Mixing of Decomposition Methods Strong Exact and Heuristic Algorithms Faster Computers Parallel Implementations Still a lot of work to do !!

29. Some Collaborators Jean-Louis Goffin Pierre Hansen Gilles Savard Marius Solomon François Soumis Jean-Philippe Vial Jean-François Cordeau Michel Denault Guy Desaulniers Daniel Villeneuve