incremental network design
play

Incremental Network Design Martin Savelsbergh Georgia Tech Aussois - PowerPoint PPT Presentation

Incremental Network Design Martin Savelsbergh Georgia Tech Aussois 2016 Collaborators Thomas Kalinowski University of Newcastle Dmytro Matsypura University of Sydney Outline Motivation Incremental Network Design with


  1. Incremental Network Design Martin Savelsbergh Georgia Tech Aussois 2016

  2. Collaborators • Thomas Kalinowski – University of Newcastle • Dmytro Matsypura – University of Sydney

  3. Outline • Motivation • Incremental Network Design with Maximum Flows – IP formulations – Heuristics – Special case: unit arc capacities • Open questions • Other results • Future research

  4. Motivation Some of my recent research projects: • Optimizing coal export chain – Transporting coal by rail from mines to terminals at a port • Optimizing carbon capture and storage – Transporting CO 2 captured at power plants via a pipeline to subsurface aquifers • Optimizing renewable energy integration – Transporting electricity produced by renewable energy sources to consumers

  5. Motivation Planning Expansions Over Time Network Network Use Operational

  6. Motivation • Where should upgrades/expansions of a network take place? • When should upgrades/expansions of a network take place? • In what sequence should upgrades/expansions of a network take place when budget and/or resources are constrained?

  7. Incremental Network Design • Network optimization problem – Minimum spanning tree, shortest path, maximum flow, minimum cost flow, TSP, … • Multiple periods – Incur the cost / realize the benefit of an optimal solution to the network optimization problem in each period – Minimize the total cost / maximize the total benefit over the planning horizon • Option to invest in capacity expansion – Limited budget to invest in capacity expansion in each period (capacity expansions may decrease cost / increase benefit)

  8. Incremental Network Design SIMPLEST VARIANT: build single edge per period at no cost

  9. Questions of interest • How difficult are incremental network design problems? • If they are difficult, are there constant factor approximation algorithms? • If they are difficult, what size instances can we solve using integer programming techniques? • If they are difficult, are there effective and efficient heuristics?

  10. Incremental Network Design with Maximum Flows • Given: – Graph D=(N, A e U A p ) , • Existing arcs and Potential arcs – Arc capacity c a – Source s – Sink t – Planning horizon T = | A p | – Option to construct one potential arc each period • Decide: – Which arc to construct in each period so as to maximize the total flow over the planning horizon (i.e., sum of the s-t flows in each period)

  11. Incremental Network Design with Maximum Flows Which path to build first? 1-1-6 = 8 1-1-1-3 = 6 0-5-6 = 11 0-0-2-3 = 5 Note: We assume that an arc built in a period can immediately be used in that period.

  12. Incremental Network Design with Maximum Flows COMPLEXITY

  13. Incremental Network Design with Maximum Flows

  14. Incremental Network Design with Maximum Flows if yes-instance 3+6+…+3n+3n+…+3n

  15. Incremental Network Design with Maximum Flows INTEGER PROGRAMMING FORMULATIONS

  16. Integer Programming flow on arc a Maximize flow out of source in period k Flow balance Arc capacity Arc capacity & forcing Budget Consistency arc a built in Initial condition or before period k Concern: Symmetry when multiple arcs need to be build to increase flow Note: Assumes T = |A p |+1 and that an arc built can be used only in subsequent periods

  17. Integer Programming An Alternative Formulation #arcs built to make a flow of f+1 possible #arcs built to make a flow of f+k+1 possible Total flow:

  18. Integer Programming Maximize total flow Flow balance Arc capacity Arc capacity & forcing Consistency Concern: Size when F – f is large

  19. Example 2 2 b a c d 1 1 Solution:

  20. Incremental Network Design with Maximum Flows HEURISTICS

  21. Heuristics STEP 1: Determine the cardinality of a minimum set of potential arcs that have to be build to increase the maximum flow from f+k to at least f+k+1 (MinArcs) capacity of arc a flow on arc a set of potential arcs build arc a or not already built

  22. Heuristics STEP 2: Among the minimal sets of potential arcs that have to be build to increase the maximum flow from f+k to at least f+k+1 choose one that increases the flow the most (MaxVal)

  23. Quickest-increment restricted set of potential arcs A p

  24. Quickest-to-ultimate Notation: r = F - f

  25. Quickest-to-target

  26. Remarks • Heuristics are not polynomial in general • Heuristics are polynomial if the maximum arc capacity of potential arcs is equal to 1 (MinArcs becomes a minimum cost flow problem) • If the maximization of the flow increment is omitted, then quickest-increment can be implemented to run in polynomial time even with general arc capacities

  27. Incremental Network Design with Maximum Flows SPECIAL CASE: All capacities equal to 1

  28. Quickest-to-ultimate path of k potential arcs Quickest-to-ultimate: 0-0-…-0-0 | 1-1-…-1-1 | 2-2 Opt: 0 | 1-1-…-1-1 | 1-1-…-1 | 2

  29. Quickest-improvement Quickest-improvement: 0-0-…-0-0 | 1-1-…-1-1 | 1-1-…-1-1 | 2 Opt: 0-0-…-0-0 | 1-1-…1-1 | 2-2-…-2

  30. Incremental Network Design with Maximum Flows SPECIAL CASE: All capacities equal to 1 and a bipartite graph

  31. Special Case Incremental Network Design with Maximum Cardinality Matchings

  32. Observations • Quickest-improvement does not always produce an optimal solution • Quickest-to-ultimate does not always produce an optimal solution • Quickest-to-ultimate is a 4/3-approximation algorithm

  33. Quickest-increment is not optimal Quickest-increment: (5,8)(7,10) (1,2)(3,4)(5,6)(9,10)(11,12)(13,14) Value: 2*5+6*6+7=53 Optimal: (1,2)(3,4)(5,6) (9,10)(11,12)(13,14) (5,8)(7,10) Value: 3*5+3*6+2*7=54

  34. Quickest-to-ultimate is not optimal Quickest-to-ultimate: (1,2)(3,4)(5,6)(7,8) (9,10)(11,12)(13,14)(15,16) Value: 4*6+4*7+2*8=68 Optimal: (1,2)(9,4) (3,4)(5,6)(7,8)(9,10)(11,12)(13,14)(15,16) Value: 2*6+7*7+8=69

  35. Incremental Network Design with Maximum Flows OPEN QUESTIONS

  36. Open Questions • Is incremental network design problem with maximum flows and capacities equal to 1 solvable in polynomial time or NP-complete? – Formulation 2 always gives integer solution – Formulation 2 is not TU • Is the incremental network design problem with maximum cardinality matchings solvable in polynomial time or NP-complete? • What are the approximation ratios of quickest-to- ultimate and quickest-improvement for the incremental network design problem with maximum cardinality matchings?

  37. Incremental Network Design with Maximum Flows OTHER RESULTS

  38. Other results • The Incremental Network Design Problem with Shortest Paths is NP-hard • The Incremental Network Design Problem with Minimum Spanning Trees is solvable in O (max{ n 2 , m log m }) time

  39. Incremental Network Design with Maximum Flows FUTURE RESEARCH

  40. • Multi-commodity flow • Investment budgets • Multi-period build times • Uncertainty

  41. QUESTIONS ?

  42. • M. Baxter, T. Elgindy, A.T. Ernst, T. Kalinowski, and M.W.P. Savelsbergh. “Incremental Network Design with Shortest Paths”. EJOR 238, 675-684, 2014. • T. Kalinowski, D. Matsypura, and M.W.P. Savelsbergh. “The Incremental Network Design Problem with Maximum Flows”, EJOR 242, 51-62, 2015. • C. Engel, T. Kalinowski, and M.W.P. Savelsbergh. “The Incremental Network Design Problem with Minimum Spanning Trees”, arXiv:1306.1926 [math.CO].

Recommend


More recommend