Intelligent vehicle scheduling for green airport baggage transport service

Abstract

Efficient airport baggage transport is critical for improving airport operation efficiency and quality. In practice, the baggage transport is usually achieved by the cooperation of tractors and trailers under the drop-and-pull mode. Recently, new electric autonomous vehicles have been introduced to promote the intelligent and sustainable development of airports. However, scheduling baggage transport vehicles presents significant challenges due to the complex relationships among tractors, trailers, and flights, which are further addressed by considering the recharging decision-making problem of electric autonomous vehicles. Besides, the airport ground handling is a highly dynamic and uncertain scenario, particularly at busy hub airports. To address these challenges, this thesis reviewed the literature related to vehicle scheduling for airport baggage transport services. Based on the previous studies and the intelligent development process of airports, this research focuses on vehicle scheduling under two operating modes: multi-trailer drop-and-pull baggage transport and electric auto-dolly-based baggage transport. For the multi-trailer drop-and-pull baggage transport, this study develops a two-stage scheduling model for tractors and trailers under the drop-and-pull mode, as well as designing an efficient hybrid intelligence-based solution algorithm. Specifically, the Adaptive Large Neighborhood Search is taken as the foundation of the algorithm, with carefully designed operators. Besides, two key methods are introduced to enhance the efficiency of the algorithm, including a K-means clustering-based initialization method and a topological sort-based solution evaluation method. For the electric auto-dolly-based baggage transport, a simplified scheduling model is established based on the model of Vehicle Routing Problem, which is then modeled into the Markov Decision Process of improvement heuristic. Then, a scheduling algorithm that integrates reinforcement learning and the Transformers variant-based deep learning model is improved, with specifically designed problem embeddings to effectively present the constraints on service time and electricity consumption, thus improving the algorithm convergence speed. Finally, supported by the flight and map data collected from real-world airports, a SUMO-based integrated airport service vehicle scheduling simulation platform is established. Simulation experimental results are analyzed to improve the algorithm and provide references for airport service vehicle scheduling in practice.