OptQuest version 7.0 includes major search efficiency improvements. We test any changes to our algorithms against a test suite over 3000 simulation optimization problems across 25 different problem types. Using an area under the curve metric which measures a normalized distance from the best known solution throughout the search, the new version 7 OptQuest is on average nearly 20% better than the latest version 6 OptQuest across all problem classes, and shows improvement across each problem class. This is a significant performance improvement. With OptQuest version 7, users will find better solutions sooner in their search. Some problem classes saw gains of greater than 40% based on the area under the curve metric.
The aggregated results for two problem classes are shown below as examples. On each graph, the x-axis shows the number of simulation runs and the y-axis shows the percentage from the best known. The blue points track the percent from best known at different numbers of simulations runs for the latest OptQuest version 6 solver engine and the yellow points show the performance of the new OptQuest version 7 solver engine. You can see that the new version has much faster convergence to the best known solution, illustrating dramatically improved performance, especially with fewer simulation runs.
Scalability up to hundreds or a few thousand parallel evaluators is now also fully supported. OptQuest internal algorithms adapt as more parallel evaluators are added, even during an active search, to productively use the extra available simulation runs in the search process. Also, the interface to add extra evaluators has been simplified, with OptQuest internally managing evaluator ids.
In addition to many internal changes, a number of minor enhancements have also been made including: