Hybridization of reinforcement learning and agent-based modeling to optimize construction planning and scheduling

Abstract

Decision-making in construction planning and scheduling is complex because of budget and resource constraints, uncertainty, and the dynamic nature of construction environments. A knowledge gap in the construction literature exists regarding decision-making frameworks with the ability to learn and propose an optimal set of solutions for construction scheduling problems, such as activity sequencing and work breakdown structure formulations under uncertainty. The objective of this paper is to propose a hybrid reinforcement learning–graph embedding network model that 1) simulates complex construction planning environments using agent-based modeling and 2) minimizes computational burdens in establishing activity sequences and work breakdown formations. Three case studies with practical construction scheduling problems were used to demonstrate applicability of the developed model. This paper contributes to the body of knowledge by proposing the hybridization of reinforcement learning and simulation approaches to optimize project durations with resource constraints and support construction practitioners in making project planning decision-making.