Abstract
Frequent order personalisation and disruptions in manufacturing are increasing intra-logistics complexity, posing greater challenges for intra-logistics Vehicle Routing Problem (VRP) resolution. VRP models are prone to structural changes, such as objective modification or constraint addition/removal, whilst knowledge reusability for model updates remains limited. Although emerging large language model (LLM) technology offers new potential for addressing this challenge, current limitations include unclear internal self-regulation mechanisms, difficulties in integrating and reusing modelling knowledge, and poor adaptability of outputs to real-world tasks. To address these issues, this study proposes a knowledge-augmented LLM framework for intra-logistics VRP optimisation modelling in digital manufacturing. It designs a self-regulating assisted modelling operational mode and a multi-step modelling knowledge augmentation method, leveraging Retrieval-Augmented Generation (RAG) and Chain-of-Thought (CoT) technologies to provide interactive intelligent assistance for intra-logistics VRP model reconstruction. Case experiment results demonstrate that the framework can efficiently reuse modelling knowledge and offer effective assistance to practitioners.
Showing the abstract — retrieve the full paper via the Exa API.