Digital twin-enabled multi-agent control for energy-efficient wood drying in desiccant-assisted heat pump systems

Abstract This article presents a digital twin-enabled framework for modeling and control of wood drying in a Desiccant-Assisted Heat Pump (DAHP) system. The digital twin integrates high-fidelity physics-based models of the heat pump, kiln chamber, and wood moisture–stress behavior, capturing coupled heat and mass transfer processes critical to drying performance. Leveraging the digital twin as the environment, the drying process is cast as a Decentralized Markov Decision Process (Dec-MDP) and addressed through Multi-Agent Reinforcement Learning (MARL). The proposed approach reduces total site energy consumption by 43.2%, shortens drying duration by approximately eight days, and decreases carbon intensity by up to 94% relative to a conventional boiler-based baseline. To enhance interpretability and support industrial deployment, a heuristic policy distilled from the MARL control achieves comparable performance. By coupling digital twin modeling with advanced learning-based control, this study establishes a deployable pathway toward sustainable, energy-efficient, and high-quality wood drying, with broader implications for next-generation smart manufacturing and energy systems.