Deep Reinforcement Learning Enabled Model for Green Last-mile Delivery in MiC Construction

With the rapid focus on a sustainable environment, carbon emission reduction has become a major issue for climate change. Construction sector, as a key industry for carbon emission reduction, mainly consumes carbon through resource delivery. The existing models for green delivery simply integrate the carbon emission factor to evaluate the performance of the forward supply chain. This paper proposes a green last-mile delivery (GLD) model based on the milk-run strategy in MiC construction, which considers the vehicle limitation and empty loaded truck delivery. The objective is to devise a strategy that minimizes the carbon emission of forward and reverse flow of delivery. The implementation of the milk-run strategy aims to collect construction waste and return it to the MiC hub from construction sites using empty trucks. This model is solved by attention-based deep reinforcement learning, utilizing the milk-run strategy. It can more accurately assess the carbon emissions during the last-mile delivery between the construction site and the MiC hub, which guides the construction and logistics companies to select a better distribution strategy under a low-carbon delivery. The results show that our proposed algorithm achieves solutions within 5% of the optimal easily and fast.