Leveraging Deep Reinforcement Learning to Facilitate the Transition from Knowledge Diversity to Knowledge Creation in Construction Project Teams

Abstract

Knowledge diversity is essential for innovation in construction project teams, yet translating it into tangible outcomes remains challenging. Traditional management theories and static methods often fail to provide dynamic solutions. This study addresses the issue through a two-stage design combining empirical analysis and deep reinforcement learning (DRL)-based simulation. Using three waves of survey data from 20 construction projects, the empirical study shows that knowledge diversity significantly predicts knowledge creation, which in turn enhances overall project performance, as reflected in core outcomes such as quality, efficiency, cost control, and timeliness. Moreover, formal and informal leadership exert contrasting moderating effects: formal leadership strengthens the benefits of functional diversity but constrains expertise diversity, while informal leadership has the opposite pattern. Building on these findings, a DRL framework based on an improved Proximal Policy Optimization (PPO)-Clip (PPO-Clip) algorithm is developed to model how leadership support can be dynamically adjusted to guide the transition from diversity to creation under resource constraints. The simulation results demonstrate that PPO-Clip not only outperforms traditional optimization methods and other DRL algorithms, but more importantly, it provides stable, efficient strategies that minimize wasted interventions and concentrate resources where they generate the greatest impact on knowledge creation. Overall, this study enriches project management research by revealing the context-dependent role of leadership in leveraging different types of knowledge diversity and by introducing a computational approach that equips managers with actionable strategies to transform diversity into sustained innovation.