Adaptive large neighborhood search for integrated planning in railroad classification yards

Abstract

Railroad classification yards serve as central hubs in single wagonload freight transportation by disassembling inbound trains and classifying outbound ones. This enables railcars to switch trains, thereby reducing the number of point-to-point connections for low demand origin-destination pairs. The quality of the operations in classification yards has a large impact on the overall performance of the system. The planning process comprises the cut generation problem, the train makeup problem, the railcar classification problem, the outbound track assignment problem, and the scheduling of service and safety operations along with the assignment of both locomotives and staff to them. This tactical planning task is nowadays mainly done manually by experienced planners and most optimization models in the literature focus only on subproblems. In the hope of filling this gap, we therefore propose a formulation for the integrated planning problem in classification yards. Since the formulation turns out to be intractable for general-purpose solvers, we propose a tailored adaptive large neighborhood search heuristic that yields high-quality results for realistic instances. Problems with up to 20 inbound and outbound trains are solved on average in less than 20 min with an average optimality gap of 0.5% for the instances for which an optimal solution is known.