Researchers Mertcan Daysalilar, Fuat Uyguroglu, Gabriel Nicolosi, and Adam Meyers from the University of California, Santa Barbara have developed a new approach to optimize electric vehicle routing, a critical challenge for sustainable logistics. Their work, published in the journal “Transportation Research Part C: Emerging Technologies,” focuses on improving the efficiency and reliability of electric vehicle fleets.
The electric vehicle routing problem with time windows (EVRPTW) is a complex task that requires minimizing travel distance, fleet size, and battery usage while adhering to strict customer time constraints. Traditional methods and even existing deep reinforcement learning (DRL) models often struggle with this problem, particularly when constraints are dense, leading to instability and poor generalization.
To address these issues, the researchers proposed a curriculum-based deep reinforcement learning (CB-DRL) framework. This framework breaks down the learning process into three phases: distance and fleet optimization (Phase A), battery management (Phase B), and the full EVRPTW (Phase C). By gradually increasing the complexity, the model can learn more effectively and stably.
The CB-DRL framework uses a modified proximal policy optimization algorithm with phase-specific hyperparameters, value and advantage clipping, and adaptive learning-rate scheduling. The policy network is built on a heterogeneous graph attention encoder enhanced by global-local attention and feature-wise linear modulation, which helps capture the distinct properties of depots, customers, and charging stations.
The model was trained on small instances with just 10 customers but demonstrated robust generalization to unseen instances ranging from 5 to 100 customers. It significantly outperformed standard baselines on medium-scale problems, achieving high feasibility rates and competitive solution quality. This approach effectively bridges the gap between neural speed and operational reliability, offering a promising solution for the energy sector’s logistics challenges.
The practical applications of this research are significant for the energy industry, particularly in optimizing the routing of electric vehicle fleets for delivery and transportation services. By improving the efficiency and reliability of these operations, companies can reduce costs, minimize environmental impact, and enhance customer satisfaction. This research highlights the potential of advanced machine learning techniques to address complex optimization problems in the energy sector.
This article is based on research available at arXiv.