📝 SummarXiv

Abstract

As a fundamental problem in transportation and operations research, the bilevel capacity expansion problem (BCEP) has been extensively studied for decades. In practice, BCEPs are commonly addressed in two stages: first, pre-select a small set of links for expansion; then, optimize their capacities. However, this sequential and separable approach can lead to suboptimal solutions as it neglects the critical interdependence between link selection and capacity allocation. In this paper, we propose to introduce a cardinality constraint into the BCEP to limit the number of expansion locations rather than fixing such locations beforehand. This allows us to search over all possible link combinations within the prescribed limit, thereby enabling the joint optimization of both expansion locations and capacity levels. The resulting cardinality-constrained BCEP (CCBCEP) is computationally challenging due to the combination of a nonconvex equilibrium constraint and a nonconvex and discontinuous cardinality constraint. To address this challenge, we develop a penalized difference-of-convex (DC) approach that transforms the original problem into a sequence of tractable subproblems by exploiting its inherent DC structure and the special properties of the cardinality constraint. We prove that the method converges to approximate Karush-Kuhn-Tucker (KKT) solutions with arbitrarily prescribed accuracy. Numerical experiments further show that the proposed approach consistently outperforms alternative methods for identifying practically feasible expansion plans investing only a few links, both in solution quality and computational efficiency.