Hybridization of evolutionary algorithm and deep reinforcement learning for multi-objective orienteering optimization

Multi-objective orienteering problems (MO-OPs) are classical multi-objective routing and have received much attention in recent decades. This study seeks to solve MO-OPs through a problem-decomposition framework, that is, an MO-OP is decomposed into knapsack problem (MOKP) traveling salesman (TSP). The MOKP TSP then solved by evolutionary algorithm (MOEA) deep reinforcement learning (DRL) method, respectively. While the MOEA module for selecting cities, DRL planning Hamiltonian path these cities. An iterative use of two modules drives population towards Pareto front MO-OPs. effectiveness proposed method compared against NSGA-II NSGA-III on various types instances. Experimental results show our performs best almost all test instances, has shown strong generalization ability.