University of Twente Student Theses
Smart refueling decisions using a reinforcement learning approach : a case study at Nijhof-Wassink
Kruit, M.N. (2024) Smart refueling decisions using a reinforcement learning approach : a case study at Nijhof-Wassink.
PDF
3MB |
Abstract: | Trucking companies face significant fuel costs, with refueling expenses accounting for approximately 30% of the operational costs. For trucking companies with low profit margins it is an ongoing challege to reduce these fuel costs. To address this challenge, this study proposes a reinforcement learning (RL) approach embedded within a sequential decision-making framework to provide near-optimal refueling policies. A unique aspect of this research is its consideration of the stochastic nature of fuel prices by integrating a machine learning prediction model into the reward function. The RL approach is validated in a case study using real-world data. In the case study the RL approach is compared against an optimal solution, a benchmark heuristic, and historical refueling decisions. Results demonstrate the effectiveness of RL, with a mere 0.3% optimality gap and a projected 11% reduction in fuel costs. In summary, this research contributes a sequential decision-making framework under price uncertainty, which defines the FRVRP as an MDP and solves it with RL. The computational results prove the contribution of an RL approach in optimizing complex refueling decisions and its potential for real-world implementation. |
Item Type: | Essay (Master) |
Faculty: | BMS: Behavioural, Management and Social Sciences |
Subject: | 01 general works |
Programme: | Industrial Design Engineering MSc (66955) |
Link to this item: | https://purl.utwente.nl/essays/98539 |
Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Repository Staff Only: item control page