In recent
literature, Zhu et al. (2011) extend the conventional SND
and include car classification and blocking, and train make-up in
a railway system. Their space–time modeling includes three layers
for service, block, and car. They design a hybrid metaheuristic algorithm
combining slope scaling, long-term memory-based perturbation
strategies, and ellipsoidal search method, which can solve
problems with up to 10 yards, 60 tracks and 3050 services. In solving
small sized instances, in case their model cannot find the optimal
solution in 10 hours, it reaches an optimality gap of 0.13%.
Moreover, in solving bigger sizes, it outperforms a commercial solver
both in time and solution quality.