Multiple optimization and segmentation technique (MOST) for large-scale bilevel life cycle optimization

Abstract

This paper introduces a new formulation for large-scale combinatorial bilevel optimization problems that involve integer, discrete, two-level decisions. The most vivid example where the new technique most applies is the life cycle optimization needed to allocate repair types and repair timings to a number of infrastructure assets (e.g., building components). Combining these decisions into a single optimization for hundreds of assets simultaneously makes the optimization problem complex and prohibitive. For such a large-scale problem, a multiple optimization and segmentation technique (MOST) is proposed to handle the optimization one level at a time through a series of small-size optimizations that can be solved easily. The performance of MOST has been validated on various problem sizes and proved to be innovative and can handle thousands of variables simultaneously.