in commercial VRP packages Evolving towards self-adaptive methods Kenneth Sörensen 1 Marc Sevaux 2 Patrick Schittekat 3 1 Centre for Industrial Management, KULeuven, Belgium, kenneth.sorensen@cib.kuleuven.be 2 University of South Brittany, France 3 University of Antwerp & Ortec, Belgium 7th EU/MEeting on Adaptive, Self-Adaptive, and Multi-Level Metaheuristics University of Málaga, Spain 16-17 November, 2006 in commercial VRP packages
1 Real-life vehicle routing 2 VNS in commercial solvers Overcoming myopic behaviour Flexibility and adaptability 3 in commercial VRP packages
Real-life vehicle routing On vehicle routing in industry Vehicle routing is very important in industry Real-life vehicle routing problems are several orders of magnitude more complex than their academic counterparts Commercial packages are popular, stand-alone or integrated To be useful, commercial packages must capture many real-life aspects This talk: Commercial vehicle routing solvers use Variable Neighbourhood Search For commercial VRP packages, adaptiveness is paramount in commercial VRP packages
Vehicle routing in theory Problem well-defined and in advance Limited set of constraints, usually only one at a time Exact methods may be used to a large extent Increasing focus on rich vehicle routing problems in commercial VRP packages
Vehicle routing problems encountered in practice Customer characteristics Driver characteristics Vehicle characteristics Route characteristics General characteristics in commercial VRP packages
Vehicle routing problems encountered in practice Customer characteristics time windows (soft/hard) pick-up/delivery/both special driver or vehicle requirements... Driver characteristics Vehicle characteristics Route characteristics General characteristics in commercial VRP packages
Vehicle routing problems encountered in practice Customer characteristics Driver characteristics hours available required resting times ability to drive some vehicles and others not legal regulations... Vehicle characteristics Route characteristics General characteristics in commercial VRP packages
Vehicle routing problems encountered in practice Customer characteristics Driver characteristics Vehicle characteristics heterogeneous fleet (different types/sizes of vehicles) multiple compartments (e.g. cooling, freezing,... ) special equipment (cranes, loading equipment,... ) varying start/end locations special licence required different cost structures... Route characteristics General characteristics in commercial VRP packages
Vehicle routing problems encountered in practice Customer characteristics Driver characteristics Vehicle characteristics Route characteristics time-dependent travel time (e.g. longer during rush hours) ability to traverse certain routes or make certain turns... General characteristics in commercial VRP packages
Vehicle routing problems encountered in practice Customer characteristics Driver characteristics Vehicle characteristics Route characteristics General characteristics non-standard different routing problems (buses, taxis, garbage collector cars, transport of disabled people,... ) multiple heterogeneous depots (e.g. carrying different products and having different stock levels for each product) stochasticity dynamic information objective function (cost, balance between route lengths,... )... in commercial VRP packages
Vehicle routing in practice in commercial VRP packages
ORMS vehicle routing survey Vehicle routing packages are used in a wide variety of environments Integration with ERP systems is important Specialization may be a good idea Size and complexity of real-life VRPs requires (meta)heuristics in commercial VRP packages
Variable Neighbourhood Search VNS in commercial solvers Overcoming myopic behaviour Flexibility and adaptability Systematically explore different neighbourhood types Increasingly succesful and popular metaheuristic Some applications to academic routing problems All commercial routing packages use it in commercial VRP packages
Variable Neighbourhood Search VNS in commercial solvers Overcoming myopic behaviour Flexibility and adaptability Systematically explore different neighbourhood types Increasingly succesful and popular metaheuristic Some applications to academic routing problems All commercial routing packages use it Why? in commercial VRP packages
VNS in commercial VRP solvers VNS in commercial solvers Overcoming myopic behaviour Flexibility and adaptability Techniques frequently used Construction heuristics (savings, clustering,...) Manipulate stops within one route Manipulate stops between routes Manipulate routes between vehicles Equalize route lengths... Techniques hardly ever used Memory structures Random perturbations Path relinking Crossover/mutation... in commercial VRP packages
VNS in commercial solvers Overcoming myopic behaviour Flexibility and adaptability VNS in commercial VRP solvers: why? Overcoming the myopic behaviour of a single neighbourhood in commercial VRP packages
VNS in commercial solvers Overcoming myopic behaviour Flexibility and adaptability VNS in commercial VRP solvers: why? Overcoming the myopic behaviour of a single neighbourhood in commercial VRP packages
VNS in commercial solvers Overcoming myopic behaviour Flexibility and adaptability VNS in commercial VRP solvers: why? Flexibility and adaptability Different problems require different approaches Developing a new method for each customer is intractable Applications can be adapted by re-arranging the neighbourhood order and frequency Some neighbourhoods operators are better suited for some problems than others Neighbourhood operators are components that may be reused Drawback: requires lots of manual intervention from expensive consultants in commercial VRP packages
Self-adaptive VRP solvers VRP solvers need to become self-adaptive Hyper-algorithm to determine optimal solver configuration One parameter: expected solution time On-line During optimization Based on actual problem data Best-suited for long optimization times Off-line Before actual optimization Based on benchmark historical data Best-suited for short optimization times Promising research area: hyper-heuristicss Important knowledge: what makes this VRP difficult? in commercial VRP packages
Conclusions Vehicle routing in industry is much more complex than in theory All commercial VRP solvers use some form of VNS Overcoming myopic behaviour of single neighbourhood Flexibility and adaptability There is a strong need for self-adaptive systems BUT: knowledge of problem structure is required What makes a specific VRP hard? Lots more research is needed in this area in commercial VRP packages
in commercial VRP packages Evolving towards self-adaptive methods Kenneth Sörensen 1 Marc Sevaux 2 Patrick Schittekat 3 1 Centre for Industrial Management, KULeuven, Belgium, kenneth.sorensen@cib.kuleuven.be 2 University of South Brittany, France 3 University of Antwerp & Ortec, Belgium 7th EU/MEeting on Adaptive, Self-Adaptive, and Multi-Level Metaheuristics University of Málaga, Spain 16-17 November, 2006 in commercial VRP packages