Using Evolutionary Algorithms for the CVRP

Context Tweaking the search heuristic based on benchmark data is a necessary evil when building and testing an OR model intended for production. But this approach has a problem: we are able to beat benchmarks but are not able to consistently get acceptable results in production, where the input data can vary for each run. One solution is to train the OR model in more benchmarks, but it is expensive for businesses to create the right benchmarks and to verify results when there’s a lot of variatio...

Submit