M.S. THESIS PRESENTATION: Genetic Algorithm Applications for the Vehicle Routing Problem with Roaming Delivery Locations

Date/Time
Date(s) - 09/07/2021
10:00

Categories No Categories


 

    Speaker: Serkan Turhan

Time: July 9, 10:00

https://zoom.us/j/99235080715?pwd=SGNLamxodVhVZk94YlRoN09XMDNiZz09
Meeting ID: 992 3508 0715
Passcode: 437504

 

Title: Genetic Algorithm Applications for the Vehicle Routing Problem with Roaming Delivery Locations

 

Abstract: The recent innovations in the e-commerce industry developed a new delivery option where the orders of the customers can be delivered to the trunks of their cars. Compared to the conventional home-delivery, this option is not only able to decrease the total distance traveled but also increase the customer satisfaction by decreasing the number of failed deliveries. The problem introduced by this option is called the vehicle routing problem with roaming delivery locations. This thesis proposes a new, time-efficient solution construction strategy for the problem. The construction strategy is able to represent any feasible solution for the problem and has a complexity linearly increasing with the number of delivery nodes in the problem. Based on the constructor, a new genetic algorithm to find and improve solutions to near-optimal within polynomial time is proposed. Furthermore, a separate, new fine-tuning algorithm to improve the parameters of the genetic algorithm for a given set of problem instances is proposed. The most notable feature of the proposed genetic algorithm is the time-efficiency as it is able to construct a solution within milliseconds for the largest problem instance available in the literature and the computation time scales with the problem size linearly. Parallel computing can be implemented in both the fine-tuning and the genetic algorithm, which will allow better results in a shorter processing time. Within 5 minutes of process time, the fine-tuned genetic algorithm found optimal solutions in 8 out of 19 instances, moreover, it was able to find solutions better than the previous solution methodologies in 12 out of 60 instances. Gaps between the results of the proposed genetic algorithm and the best solution found by the solver are between (0.0%, 26.2%) and (-6.0%, 16.3%) in small-medium and large instances respectively.