SplitNet: A Reinforcement Learning Based Sequence Splitting Method for the MinMax Multiple Travelling Salesman Problem

MinMax Multiple Travelling Salesman Problem (mTSP) is an important class of combinatorial optimization problems with many practical applications, which the goal to minimize longest tour all vehicles. Due its high computational complexity, existing methods for solving this problem cannot obtain a solution satisfactory quality fast speed, especially when scale large. In paper, we propose learning-based method named SplitNet transform single TSP solutions into mTSP same instances. Specifically, generate sequences and split them subsequences using attention-based model trained by reinforcement learning. We also design decision region splitting policy, significantly reduces policy action space on instances various scales thus improves generalization ability SplitNet. The experimental results show that generalizes well outperforms baselines Google OR-Tools widely-used random datasets different public speed.