Abstract
Dynamic scheduling in real-world environments often struggles to adapt tounforeseen disruptions, making traditional static scheduling methods andhuman-designed heuristics inadequate. This paper introduces an innovativeapproach that combines Genetic Programming (GP) with a Transformer trainedthrough Reinforcement Learning (GPRT), specifically designed to tackle thecomplexities of dynamic scheduling scenarios. GPRT leverages the Transformer torefine heuristics generated by GP while also seeding and guiding the evolutionof GP. This dual functionality enhances the adaptability and effectiveness ofthe scheduling heuristics, enabling them to better respond to the dynamicnature of real-world tasks. The efficacy of this integrated approach isdemonstrated through a practical application in container terminal truckscheduling, where the GPRT method outperforms traditional GP, standaloneTransformer methods, and other state-of-the-art competitors. The keycontribution of this research is the development of the GPRT method, whichshowcases a novel combination of GP and Reinforcement Learning (RL) to producerobust and efficient scheduling solutions. Importantly, GPRT is not limited tocontainer port truck scheduling; it offers a versatile framework applicable tovarious dynamic scheduling challenges. Its practicality, coupled with itsinterpretability and ease of modification, makes it a valuable tool for diversereal-world scenarios.