Enhancing University Building Energy Flexibility Performance Using Reinforcement Learning Control

The Building Energy Flexibility (BEF) system has the potential to reduce carbon emissions and energy consumption by integrating flexible loads and renewable energy sources. However, without proper optimization, BEF implementation can lead to demand-supply imbalances (DSI), causing voltage fluctuations and power quality issues. This study focuses on addressing these challenges by implementing a flexible load control (FLC) based on reinforcement learning (RL), particularly aimed to optimizing BEF performance in a university building setting. The case study focuses on the LABTEK XIX building in Bandung City, Indonesia, which comprises various zones, including offices, lobbies, libraries, classrooms, and laboratories. Each zone has distinct schedules and activities, resulting in significant variations in energy demands and occupancy patterns. The BEF system was modeled using OpenStudio and EnergyPlus, providing a training environment for the controller. The controller was developed in Python using the OpenAI Gym framework and leverages the Proximal Policy Optimization (PPO) algorithm to manage Heating, Ventilation, and Air Conditioning (HVAC) systems and local photovoltaic (PV) energy supply. By addressing the dynamic nature of flexible loads and the intermittent supply of PV energy, the controller enhances system resilience against disturbances like weather variations and occupancy changes. The key performance indicators (KPIs) evaluated include increasing building self-consumption, energy flexibility, and energy savings while maintaining thermal zones within acceptable comfort levels for occupants. It was evaluated over four weeks during the equinoxes and solstices. The controller successfully increased average self-consumption by 6.58%, achieved an average energy flexibility of 4.91 kWh, and realized energy savings up to 3 kWh, all while ensuring thermal comfort in the managed zones.