Experimental data and the executable to test the performance of situation-adaptive policy for container stacking in an automated container terminal
Description
Determining where to stack the containers at the storage yard of a container terminal is an important problem because that decision critically affects the efficiency of container handling in the yard and eventually the efficiency of the vessel operations that is considered the most important for the productivity of the whole terminal. One limitation of the stacking policies previously proposed is that they are static in nature. Although good locations for stacking may change as the workload of vessel operation changes, the previous policies are insensitive to such changes. Failure to recommend good locations leads to elongated operations of yard cranes and thus makes them hard to keep up with the workload of vessel operation. In this paper, we propose a method for deriving a dynamic policy that can adapt to the workload of vessel operation that changes over time. Our method derives two boundary polices: one for very high workload and the other for very low. Then, a policy appropriate for any intermediate workload can be synthesized from the two boundary policies through an interpolation. Simulation experiments showed that the proposed policy significantly reduced overall container handling time compared to the previous static policy. When measured in terms of the time the transportation vehicles wait for container handling services, the improvement was about 19%.
Files
Steps to reproduce
After unzip the experiment.zip, there are three directories: Dynamic, Static, and Scenario. - The Scenario directory contains 1000 scenario data for training and 100 scenario data for testing. - If you run the run.exe file in the Dynamic directory, a dynamic policy is optimized and 100 tests for the dynamic policy are performed. - If you run the run.exe file in the Static directory, a static policy is optimized and 100 tests for the static policy are performed. * The program can be run on Windows. You may need to install the VC++ 12.0 redistributable package.