AI utilization in route planning for delivery trucks within the supply chain

Sammanfattning: The supply chain has potential for growth. Many different parts can be optimized, and every possible improvement has not yet been tested. This study focuses on route optimization with the specific goal of figuring out how machine learning can be applied to route planning and how key factors that impact travel time for a route can be taken into account for this problem. Time was dedicated to learning about machine learning, how it is applicable to route planning, as well as potential key factors that can be used with different machine learning algorithms. Once enough knowledge had been gathered, a prototype was implemented to verify key factors usability in route planning and test different route planning problems, such as point-to-point routes and traveling salesman problem. Key factors were gathered during the thesis work, and based on the result of the thesis, their ability to be used in route optimization was verified. Methods of collecting the key factors were looked into, and two algorithms were tested that had the potential of using these factors. The two algorithms proved the usability of key factors and showed their potential in route planning problems. First, the “Neural Evolution of Augmenting Topologies” algorithm was tested and verified that it could solve simple route planning problems. Although, it was later overshadowed by a genetic algorithm solution, which could solve point-to-point travel better and showed usefulness in the traveling salesman problem. The thesis work did not provide a full-scale solution to optimizing route planning. However, several conclusions were made on the topic, such as the possibility of training neural networks using supervised learning to calculate edge cost and genetic algorithms showing its potential in multi-stop route planning. We believe that several of the conclusions made in the thesis could show promise in the area of route optimization given enough resources.