In the rolling horizon problems, while these still did well, they could not maintain their dominance.
Many of the other algorithms provided better results on some of the problems.
The difference between the four Urban’s method based algorithms was not large.
At the same time the algorithms that did poorly in the large fixed period problems appear to do even worse on the rolling horizon problems.
These were the DP based algorithms that did not have the ability to self-adjust to the changing data.
This underlines the importance of using algorithms that have some self-adjusting capability when rolling horizons are considered.
Also our experiments demonstrate the importance of developing quick and effective heuristics for the dynamic layout problem under rolling horizons since it appears that no one algorithm is able to do well in all situations.
Thus we cannot rely on one algorithm to consistently give good results.