There is a bug in the modelling of Scheduling Benchmark instance 15. In this article, we will report on the problem and explain how to analyze the bug.
Modelling bug:
In the coverage constraint section, there is a bug where a soft constraint with a Max value of 0 is interpreted as a hard constraint.
<Bug section>
Marked portions should be as follows.
<Load instance15.ros>
As shown in the figure below, you can see that the Max 0 part of the coverage constraint is actually a hard constraint, when it should be a soft constraint with a weight of 1.
<Bug workaround>
Change them to soft constraints with a weight index of 1, as shown in the figure below.
<Checking the bug fix>
Export the schedule to Excel as follows.
<Replace with UB3829 solution>
Using the format above, replace the data to UB3829 solution (text format) which was obtained by asking the administrator of the benchmark site.
<Import the UB3829 solution above in another Schedule Nurse>
<Lock all schedules on the first Schedule Nurse>
We preserve the scheduling soft constraints as is.
<Copy and paste the second Scheduler solution>
<Lock all the schedule on the first Schedule Nurse>
<Fill with Day Offs on the schedules in the first Schedule Nurse>
<Solve it!>
We obtained the repeated result of UB3829.
If you use Algorithm 1, please take care allowable errors on column constraints as follows.
<Summary>
Incorrect modeling causes missing the 3829UB solution. In this state, we never get the 3829UB. After workarounds, we obtained the repeated result of 3829UB.
0 件のコメント:
コメントを投稿