A new article from our research assistant Ali Koç!
Tarih: Date
The study presents a new Reinforcement Learning-assisted Genetic Algorithm-based solution framework for the multi-objective capacitated vehicle routing problem with time windows. The proposed approach aims to reduce the total route duration and the number of vehicles used, while increasing the satisfaction obtained from customers served within their time windows. Through the Q-learning-based operator selection mechanism integrated into the NSGA-II framework, crossover and mutation operators are selected adaptively according to the current performance of the solutions. Five different Q-learning-based operator selection strategies were compared with fixed and random operator selection approaches; the statistical analyses showed that the learning-based approaches performed better than the fixed-operator approach, while the random selection strategy was outperformed by most of the learning-based methods. In addition, identifying the most suitable operators for each solution state and using them within NSGA-II provided the best overall performance. In this respect, the study proposes an adaptive operator selection approach that produces competitive results in terms of Pareto-front quality and guides the evolutionary search process more effectively.
You can access the related article from the link.