Enhancing Energy Efficiency and Flexibility in Educational Buildings Through a Deep Reinforcement Learning-Based Controller for Rooftop Units

Advanced controllers based on predictive and adaptive frameworks play a pivotal role in the optimisation of building energy management and in the exploitation of building energy flexibility. This study analyses the application of a deep reinforcement learning controller tailored for managing four RoofTop Units (RTUs) serving 4 classrooms in an educational building at Politecnico di Torino Campus, equipped with solar PV panels and a battery storage system. The controller was designed to reduce energy costs while maintaining desirable indoor temperature conditions. The integration of renewable energy sources and storage, provide a holistic approach to the management of integrated energy systems. The control agent was initially trained on a single RTU serving a classroom and successively transferred and deployed on the remaining systems. After the transfer process, the proposed controllers effectively manage each RTU and their interactions, leveraging energy generation and storage to reduce reliance on the grid. To assess the performance of the developed controller, a simulation environment combining Modelica and Python was employed. The simulation results highlight the proposed controller's ability to be effectively transferred among similar systems while achieving improvements in energy management. Compared to a traditional control strategy, the proposed solution effectively reduces operational costs while maintaining comfort standards within the building environment. The results obtained demonstrate the potential of deep reinforcement learning strategy in enhancing building energy management and underscore its effectiveness in increasing the flexibility of integrated energy systems in buildings.