Ing. Vilém Heinz

Ing. Vilém Heinz

Ing. Daniel Vašata, Ph.D.

Department of Applied Mathematics

In this work we use Transformer architecture to find a solution for the Flow Shop Scheduling Problem, which we can then use to accelerate a constra- int programming solvers search for optimal solutions. We base our model on the architecture of TSP Transformer, which uses Transformers for solving the Travelling Salesman Problem. We sucessfully adapt the architecture to the Permutation Flow Shop Scheduling Problem and use the trained model for generating warmstart for the IBM CP Optimizer solver. This accelerates the solvers search for optimal solutions by an average of 144%, relatively to the searches not utilizing warmstart. We further generalize the architecture to the Nonpermutation Flow Shop Scheduling Problem. Due to its increased complexity however, the models training process becomes more challenging, and the model does not achieve adequate performance. Thus, addressing the Nonpermutation Flow Shop Sche- duling Problem requires deeper alterations to the models architecture.