Using Column Generation for the Pickup and Delivery Problem with Disturbances

This thesis shows how both the PDPTW (Pickup and Delivery Problem with Time Windows) and a variant where travel times are non-deterministic (Pickup and Delivery Problem with Disturbances), can be solved by an approximation algorithm that uses Linear Programming with column generation. In the pricing problem – which is solved by Simulated Annealing – an Evolutionary Algorithm helps to execute better performing sequences of mutations more often. Before making the Linear Program integer, some requests will temporarily be removed to create columns that helps to find that integer solution easier. This strategy outperforms existing algorithms on some of the PDPTW benchmarks. When the disturbances are taken into account during computation, the solution quality improves much compared to algorithms that assume deterministic travel times on problems that are nondeterministic in reality.