Adaptive Crossover Operators in Memetic Algorithms for Solving the Capacitated Vehicle Routing Problem

Optimizing delivery routes remains a fundamental challenge in logistics, where the Capacitated Vehicle Routing Problem (CVRP) serves as a key optimization framework. This paper introduces an enhanced memetic algorithm, building upon Vidal's Hybrid Genetic Search, that incorporates a novel family of Adaptive Crossover (AX) strategies. These strategies dynamically adjust recombination behavior based on real-time search performance and solution quality feedback. Extensive experiments on the standard CVRP benchmarks from Uchoa et al. (2017) demonstrate that the best AX configuration reduces the average optimality gap by 0.89 percentage points compared to the classical Order Crossover (OX), representing an 18.7\% relative improvement. Our findings establish adaptive recombination as a powerful mechanism for enhancing both solution quality and convergence efficiency in vehicle routing optimization.