Adaptive Scheduling In Dynamic Environments: A Data-Driven Approach With Digital Twins And Deep Reinforcement Learning

We propose a two-phase data-driven approach that incorporates Digital Twins (DTs) to capture and represent real-time data and leverages Deep Reinforcement Learning (DRL) to learn optimal solutions within a Markov decision process framework. We address the challenges of solving complex, real-life large-scale scheduling problems in dynamic environments which are computationally expensive and require approximations and heuristics to solve. Additionally, we represent the underlying environment as a graph and employ Graph Neural Networks (GNNs) for efficient data processing. Our proposed approach offers a promising solution to enhancing decision-making in dynamic and uncertain environments across various operational scheduling domains, ensuring efficient resource allocation even in the face of real-time uncertainties.