IBPSA Project 2 Building Optimization Testing Framework (BOPTEST): Overview, new software features, and example use cases

Demand-side flexibility (DSF) is defined as ‘the capability of any active customer to react to external signals and adjust their energy generation and consumption in a dynamic time-dependent way, individually as well as through aggregation.’ Given rising electricity prices and buildings’ significant role in global emissions, DSF is crucial. It helps strategically adjust building energy use during peak demand or high-price periods, optimizing consumption and reducing costs.

This study utilizes model-free Reinforcement Learning (RL) to optimize Heating, Ventilation, and Air Conditioning (HVAC) operations in response to dynamic market conditions. The approach integrates a Soft Actor-Critic agent with a tailored reward function that combines an environment-adaptive thermal reward, a consumption-optimized energy reward, and a Time-of-Use tariff-based weighting factor. By adapting to fluctuating prices, varying occupancy states, and changing outdoor environments, the method dynamically balances thermal comfort and energy savings, effectively reducing energy expenditure and enhancing DSF. Validation uses two open-source simulation platforms. The first, the Building Optimization Testing (BOPTEST) framework, is a physics-based simulator equipped with standardized Key Performance Indicators (KPIs). Leveraging BOPTEST’s standardized test cases, we benchmark our method against three advanced strategies: an optimized, price-unaware rule-based controller embedded in BOPTEST, and two price-aware RL-based approaches that utilize BOPTEST’s KPIs as reward references ^[1]. According to the KPIs, our method, alongside two other RL-based controllers, achieves satisfactory thermal comfort and significantly outperforms the rule-based approach. Moreover, our method reduces energy usage by up to 29% and operational costs by up to 30% compared to other price-aware RL-based alternatives. To further quantify these improvements, we propose a new Flexibility Index (FI) offering a holistic assessment that not only quantifies flexibility but also emphasizes thermal comfort satisfaction. Results show a 38% increase in flexibility compared to the price-unaware method and superior performance relative to other price-aware RL-based controllers. The second platform, a data-driven HVAC simulation model, acts as a digital twin using machine learning and real-world data to simulate responses to varied inputs ^[2].

The successful application of our strategy on this platform confirms its adaptability to real-world implementations. Our method consistently delivers superior performance across a wide range of conditions—from physics-based to data-driven simulation platforms, encompassing diverse climates from hot to cold, different building typologies, and various HVAC system configurations and control mechanisms. While guaranteeing thermal comfort, it reduces operational costs compared to other advanced price-aware controllers, resulting in a significant increase in DSF.