Multi-objective Adaptive Dynamics Attention Model to Solve Multi-objective Vehicle Routing Problem

Multi-objective combinatorial optimization problems (MOCOP) are commonly encountered in everyday life. However, finding the optimal solution through traditional exact and heuristic algorithms can be time-consuming due to its NP-hard nature. Fortunately, deep reinforcement learning (DRL) has shown promise in solving complex combinatorial optimization problems (COP). In this paper, we introduce a new Multi-objective Adaptive Dynamics Attention Model (MOADAM) that aims to better approximate the whole Pareto set. We modify the encoder and decoder of the model to better utilize dynamic information, and we also design a new weight sampling method to improve the model’s performance for extreme solutions. Our experimental results demonstrate that our proposed model outperforms the current state-of-the-art algorithm in terms of solution quality on multi-objective vehicle routing problems with capacity constraints (MOCVRP).