Optimization of Train Schedules

Abstract

Logistics and transportation are extremely important to modern society. In this thesis we investigate how to optimize the pathing and scheduling of trains, particularly targeting automated mining transports, what we call Path Conflict Resolution (PCR). A restriction imposed is that solutions to this problem must be deadlock-free, which implies that trains may not be treated separately but must be scheduled together. We approach this problem using a combination of Alternative Graphs, Mixed Integer Linear Programming (MILP), and Variable Neighbourhood Search (VNS), building upon previous work on the scheduling of passenger trains. Our model extends the previous work by including additional complications in the scheduling, different objective functions, modelling of train lengths and moving blocks, and improvements in the MILP problem generation. We successfully schedule trains on a large realistic rail network within acceptable computation times. The strongest improvement is the development of couplings in the alternative graph, which in our experiments leads to a tenfold reduction in the number of binary variables.