A general deep reinforcement learning hyperheuristic framework for solving combinatorial optimization problems

Many problem-specific heuristic frameworks have been developed to solve combinatorial optimization problems, but these do not generalize well other problem domains. Metaheuristic aim be more generalizable compared traditional heuristics, however their performances suffer from poor selection of low-level heuristics (operators) during the search process. An example in a metaheuristic framework is adaptive layer popular Adaptive Large Neighborhood Search (ALNS). Here, we propose hyperheuristic that uses Deep Reinforcement Learning (Deep RL) as an alternative ALNS. Unlike which only considers heuristics’ past performance for future selection, RL agent able take into account additional information process, e.g., difference objective value between iterations, make better decisions. This due representation power methods and decision making capability can learn adapt different problems instance characteristics. In this paper, by integrating ALNS framework, introduce Hyperheuristic (DRLH), general solving wide variety show our at selecting each step process Uniform Random Selection (URS). Our experiments also while properly handle large pool DRLH negatively affected increasing number heuristics.