The data of the article "Combining Deep Reinforcement Learning with Heuristics for Solving the Traveling Salesman Problem"

We have developed a new framework for learning improvement heuristics, which automatically discovers better improvement policies for heuristics to iteratively solve the TSP. Our framework proposes a novel approach (named RL-SA), which combines deep RL with SA algorithm, to improve 2-opt heuristic algorithm’s performance in learning a pairwise selected policy of next promising solutions. The novelty of the RL-SA approach also lies in that it leverages the Whale Optimization Algorithm (WOA) to generate an initial solution for better sampling efficiency, and that it adopts the Gaussian perturbation strategy for perturbing the state to tackle the sparse reward problem of the RL algorithm. Our program achieved excellent performance in solving the Traveling Salesman Problem.