A Q‐Learning Driven <scp>NSGA</scp> ‐ <scp>II</scp> for Vehicle Routing Problem With Service Quality and Resource Balance

ABSTRACT In urban delivery operations, service quality plays a key role in determining enterprise competitiveness, while balanced resource allocation is crucial for operational efficiency. However, few studies have investigated the factors that influence service quality and resource balance, or the trade‐offs between them. To fill this gap, this paper proposes a vehicle routing problem considering service quality and resource balance (VRPSQRB), in which stochastic quality rewards (SQR) capture the uncertain gains from differentiated service performance and develops a corresponding multiobjective mathematical model. Since the VRP is known to be an NP‐hard problem, solving it with three simultaneous objectives, namely cost, service quality, and resource balance, further increases its complexity. To address this challenge, we propose a Q‐learning driven NSGA‐II (QNSGA‐II) solution framework. In QNSGA‐II, a state representation tailored to the VRPSQRB is designed to help balance the three conflicting objectives. A hybrid strategy library, combining genetic and local search methods suited to the characteristics of VRPSQRB, is also developed. Q‐learning is used to select optimal combinations of strategies based on a multidimensional population state, enabling adaptive control beyond traditional hyperheuristic frameworks. Finally, QNSGA‐II is evaluated using Solomon benchmark datasets with customer sizes of 25, 50, and 100. The experimental results show that QNSGA‐II achieves average improvements of 11.46% in hypervolume (HV) and 20.20% in inverted generational distance (IGD) compared to baseline algorithms, demonstrating its efficiency and robustness.