Sustainable Supply Chain Strategies for Modular-Integrated Construction Using a Hybrid Multi-Agent–Deep Learning Approach

Abstract

Modular integrated construction (MiC) is a cutting-edge approach to construction that significantly improves efficiency and reduces project timelines by prefabricating entire building modules off-site. Despite the operational benefits of MiC, the carbon footprint of its extensive supply chain remains understudied. This study develops a hybrid approach that combines multi-agent simulation (MAS) with deep learning to provide scenario-based estimations of CO2 emissions, costs, and schedule performance for MiC supply chain. First, we build an MAS model of the MiC supply chain in AnyLogic, representing suppliers, the prefabrication plant, road transport fleets, and the destination site as autonomous agents. Each agent incorporates activity data and emission factors specific to the process. This enables us to translate each movement, including prefabricated components of construction deliveries, module transfers, and module assembly, into kilograms of CO2 equivalent. We generate 23,000 scenarios for vehicle allocations using the multi-agent model and estimate three key performance indicators (KPIs): cumulative carbon footprint, logistics cost, and project completion time. Then, we train artificial neural network and statistical regression machine learning algorithms to captures the non-linear interactions between fleet allocation decisions and project outcomes. Once trained, the models are used to determine optimal fleet allocation strategies that minimize the carbon footprint, the completion time, and the total cost. The approach can be readily adapted to different MiC configurations and can be extended to include supply chain, production, and assembly disruptions.