Adaptive Parameter Setting for Genetic Algorithms Using Reinforcement Learning: A Case Study on the Capacitated Vehicle Routing Problem

Abstract

This paper presents an innovative approach that integrates reinforcement learning (RL) with genetic algorithms (GA) to adaptively optimize parameters for solving the Capacitated Vehicle Routing Problem (CVRP). Traditional static approaches, such as Design of Experiments (DOE), often struggle to maintain diversity within the population pool, leading to suboptimal solutions. The proposed RL-GA method dynamically adjusts the GA parameters, resulting in improved solution quality across a set of benchmark CVRP problems. The RL-GA not only outperforms static methods but also demonstrates potential for broader application in other combinatorial and nonlinear optimization problems. Future work includes testing the RL-GA on larger CVRP instances, which, if successful, could significantly enhance the efficiency of solving complex practical problems using genetic algorithms.