Deep Reinforcement Learning-Based Real-Time Controller for Energy-Efficient Buildings

Abstract

Energy-efficient buildings play an important role in driving the transition toward more efficient and sustainable energy systems, especially in green local energy communities. However, incorporating such buildings into larger energy networks poses considerable difficulties, particularly in developing control systems that are both effective and efficient. In this paper, a deep reinforcement learning (DRL) approach is proposed to optimize the control of various decision variables within an integrated energy system including an energy-efficient building. The Markov decision process is formulated to maximize the building's profit from energy transactions with the electricity grid while simultaneously maintaining indoor temperatures at levels preferred by the occupants. To achieve this, the heating, ventilation, and air conditioning (HVAC) system is scheduled using the DRL method. Specifically, a soft actor-critic (SAC) agent is trained to manage an energy system including a real-case energy-efficient building in Lahti city of Finland with HVAC control system, an energy storage system, solar panels, and energy interactions with the grid. The results demonstrate the ability of the SAC agent to learn near-optimal decision-making strategies, increasing the economic performance of while ensuring thermal comfort for residents. This approach highlights the potential of DRL in enhancing both economic and environmental outcomes in energy building management.