11.07.2015 Views

View - Universidad de Almería

View - Universidad de Almería

View - Universidad de Almería

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

228 J. Cole Smith, Fransisca Sudargho, and Churlzu LimIn or<strong>de</strong>r to force the w-variables to match the enemy’s true solution, we can enact the followingcutting plane method. Suppose that we solve (1) and obtain variables ˆx and ŵ. Letw i be the true action of the enemy for i ∈ A. For each i ∈ A such that w i < ŵ i , we add thefollowing constraint:∑i−1w i ≤ w i x i + (min{1 − w i , b g /b i }) (1 − ˆx g )x g . (2)g=1Each such constraint is then passed back to (1), and the mo<strong>de</strong>l is resolved until w = ŵ. Essentially,(2) imposes an upper bound on w i of w i , unless some other higher-preference arcsthat are not currently being built will be built in the next iteration, thus forcing the enemy toexpend resources on other arcs. There exist several other necessary conditions for w-feasibilitythat can be stated a priori in (1), but we omit them here for the sake of brevity.A mo<strong>de</strong>l that can be solved statically makes use of the fact that there will exist only one w-variable value that can be fractional, as in the case in any knapsack <strong>de</strong>cision problem. Definebinary <strong>de</strong>cision variables z i ∀i ∈ A equal to one if and only if x i = 1, all constructed arcswith an in<strong>de</strong>x smaller than i are completely <strong>de</strong>stroyed, and all constructed arcs with an in<strong>de</strong>xgreater than i are not affected by the enemy. Arc i itself may be interdicted in any manner.The total enemy interdiction must exhaust their entire budget. (This forces us to build atleast enough capacity so that the enemy can <strong>de</strong>stroy B units; we can remove this restrictiveassumption by use of some simple mo<strong>de</strong>ling tricks.) These restrictions are enforced as follows:∑z i = 1g∈Aw i ≤(3a)i∑z g ∀i ∈ A (3b)g=1|A|∑w i ≥ x i − z g ∀i ∈ A (3c)g=iw i ≤ x i ∀i ∈ A (3d)∑b i (x i − w i ) = B.(3e)i∈AOur preliminary computational experience has shown that incorporating (3) into (1), and solvingas a static integer program, is a much more effective method of solving this problem thanby using the initial cutting-plane method <strong>de</strong>veloped above.2.2 Flow-Based Greedy Interdiction CaseIn this case, the enemy wishes to interdict the arcs having the most initial flows on them. Sincethe enemy <strong>de</strong>cision does not <strong>de</strong>pend on a simple set of binary <strong>de</strong>cision variables, the foregoingalgorithm must be modified for this case.Once again, we <strong>de</strong>termine an in<strong>de</strong>x i ∗ such that arc i ∗ may be partially <strong>de</strong>stroyed, implyingthat every arc with a greater flow than i ∗ must be completely <strong>de</strong>stroyed, while all arcs witha smaller flow than i ∗ cannot be interdicted. Unlike the previous case, we do not <strong>de</strong>fine a<strong>de</strong>cision variable to <strong>de</strong>termine the i<strong>de</strong>ntity of i ∗ , but must solve one integer program for eachpossible value that i ∗ can take. If there exists an arc that has the same amount of flow as i ∗ , wewill use the enemy’s secondary priority to <strong>de</strong>termine whether or not it is <strong>de</strong>stroyed. Lettingp i ∀i ∈ A be the enemy’s secondary priority for arc i (a smaller number <strong>de</strong>notes a higher

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!