| This thesis studies a scheduled service network design problem for rail freight transportation planning. Rails follow a special two level consolidation organization, and the car-to-block, block-to-service handling procedure complicates daily operations. In this research, the two consolidation processes as well as the operation schedule are considered simultaneously, and by solving this problem, we provide an overall cost-effective operating plan, including blocking policy, train routing, scheduling, make-up policy and traffic distribution.;The proposed model relies on a complex mixed-integer programming formulation. The problem is very hard to solve due to the computational difficulty as well as the tremendous size of the application instances. Three versions of the problem are studied, which are the simplified model (with only non-stop services), complete model (with both non-stop and multi-stop services) and very-large-scale complete model. Heuristic algorithms are developed to provide good feasible solutions in reasonable computing efforts.;A special case with non-stop services is first studied. According to a specific characteristic of the direct service network design problem, we develop a tabu search algorithm. The tabu search moves in a cycle-based neighborhood, where flows on blocks are re-distributed according to the cycles in a conceptual residual network.;A slope scaling based algorithm is developed for the complete model, and we propose a new method, called ellipsoidal search, to further improve the solution quality. Ellipsoidal search combines the good feasible solutions generated from the slope scaling, and collects the features of good solutions into an elite problem, and solves it with exact solvers. The algorithm thus takes advantage of the convergence speed of slope scaling and solution quality of ellipsoidal search, and is proven effective. The algorithm also presents an alternative for solving the simplified problem.;In order to describe various rail operations at the tactical level, we extend the physical network and construct a 3-layer time-space structure, in which the time dimension takes into consideration the temporal impacts on operations. Furthermore, operations on trains, blocks, and cars are described in different layers. Based on this network structure, we model the rail planning problem to a service network design formulation.;Finally, we work on the very-large-size complete model. A hybrid heuristic is developed by integrating the ideas of previous research with column generation. We propose a new slope scaling scheme where, compared with the previous scheme, only approximate service costs instead of both service and block costs are considered. The new slope scaling scheme thus separates the block decisions and service decisions, and provide a natural decomposition of the problem. Experiments show the algorithm is good to solve real-life size instances.;Keywords: service network design, rail freight transportation, time-dependent network design. |
