サプライチェーン再設計問題における探索実績情報を利用した分枝探索アルゴリズム

抄録

Mixed-integer programming (MIP) is a key-technology for the optimization of relevant business processes such as supply chain network design. Many practical aspects like demand variations often require the solution of a sequence of very similar MIPs, in which two consecutive problems share most of their data. Using a state-of-the-art MIP solver for every model in the sequence, branching information from previous models is usually lost and needs to be reinitialized during an expensive strong-branching procedure. In this paper, we investigate the benefits of transferring branching information from previous runs when sequences of similar models are solved. Computational experiments confirm that the proposed method effectively reduces the overall solving time on a set of instances from both MIPLIB2003 and real-world scenarios