Lab
Building Automation and Energy Data Analytics Lab
Institution: Polytechnic University of Turin
Department: DENERG - Department of Energy
About the lab
BAEDA is a research lab in TEBE group aimed at contributing in bridging the gap between building physics and data science supporting the transition toward novel paradigms of energy management in buildings and energy grids.
www.baeda.polito.it
www.baeda.polito.it
Featured research (60)
In recent years, Transfer Learning (TL) has emerged as a promising solution to scale Deep Reinforcement Learning (DRL) controllers for building energy management, addressing challenges related to DRL implementation as high data requirements and reliance on surrogate models. Moreover, most TL applications are limited to simulations, not revealing their real performance in actual buildings. This paper explores the implementation of an online TL methodology combining imitation learning and fine-tuning to transfer a DRL controller between two real office environments.
Pre-trained in simulation using a calibrated digital twin, the DRL controller reduces energy consumption and improves indoor temperature control when managing the operation of a Thermally Activated Building System in one of the two offices both in simulation and in the real field. Afterwards, the DRL controller is transferred to the other office following the online TL methodology. The proposed approach outperforms a DRL controller implemented without pre-training, and Rule-Based and Proportional-Integral controllers, achieving energy savings between 6 and 40% and improving indoor temperature control between 30 and 50%. These findings underscore the efficacy of the online TL methodology as a viable solution to enhance the scalability of DRL controllers in real buildings.
The paper introduces a novel methodology for optimizing the operation of a centralized Air Handling Unit (AHU) in a multi-zone building served by VAV boxes with interpretable rules extracted from a Deep Reinforcement Learning (DRL) controller trained to enhance energy efficiency and indoor temperature control. To ensure practical application, a Rule Extraction (RE) framework is developed, translating the DRL complex decision-making process into actionable rules using decision trees. A multi-action approach is proposed by developing three different regression trees for adjusting the supply water temperature, the position of the chiller valve, and the position of the economizer damper of the AHU. The extracted rules are benchmarked against the original DRL controller and two conventional control sequences based on ASHRAE 2006 and ASHRAE Guideline 36 within a high-fidelity co-simulation architecture combining Spawn of EnergyPlus and Python. The co-simulation environment uses EnergyPlus for building envelope and loads while HVAC components and controls are implemented in the equation-based modeling language Modelica. Results show that the RE-based controller closely approximates the performance of the DRL policy with an electric energy consumption only 3% higher, highlighting its ability to effectively mirror a more complex control logic, representing a transparent and easily implementable alternative. The controllers based on ASHRAE 2006 and ASHRAE Guideline 36 lead to higher energy consumption (for both chiller and fan) and violations of indoor temperature compared to both RE-based control and DRL. This study underscores the potential of integrating AI-driven control methods with interpretable rule-based systems, facilitating the adoption of advanced energy management strategies in real-world building automation systems.
With the widespread adoption of renewable energy systems in residential buildings, particularly in the context of collective self-consumption groups (CSC) and Renewable Energy Communities (REC), understanding user behavior becomes pivotal for enhancing energy efficiency and increasing the energy share among participants for an optimal use of renewable resources. Regardless of which configuration is adopted (CSC or REC), a key aspect is how to share the generated economic benefits from the self-produced energy and identify the fairest way to distribute the incentive derived from the shared energy among users.
In this context, the aim of this work is to introduce a data-driven energy benchmarking process that leverages the analysis of long-term monitoring data of residential buildings to i) characterize energy consumption patterns of users over time, ii) support the development of an optimal incentive sharing mechanism among users involved in such legal entities. The proposed approach is tested on a monitored residential building, located in Northern Italy, which includes 13 flats and is equipped with a centralized photovoltaic system.
Renewable Energy Communities (REC) can largely contribute to building decarbonization targets and provide flexibility through the adoption of advanced control strategies of the energy systems. This work investigates how the role of flexibility sources will be impacted by shifting towards advanced control strategies under a high penetration of variable Renewable Energy Sources, in the following years. A large residential area with diverse energy systems, building envelope configurations, and energy demand patterns is modeled with the simulation environment RECsim, a virtual testbed for the implementation of energy management strategies in REC. Photovoltaic (PV) panels, Battery Energy Storage and Thermal Energy Storage (TES) of different sizes for each household provide a realistic description of a REC which includes both consumers and prosumers.
This study explores a scenario in which advanced controllers based on Deep Reinforcement Learning (DRL) replace existing Rule-Based Controllers in building energy systems across a significant number of buildings. These control policies are simulated under three different scenarios that consider consumers with different pricing schemes and TES penetration.
Efficient control strategies, have demonstrated significant potential, regardless of the presence of thermal storage and ToU pricing schemes, in reducing energy demand by 12.6%, cutting energy costs by 20.8%, and enhancing self-sufficiency and self-consumption, with minimal impact on Shared Energy. Implementing a flat tariff scheme under DRL enables consumers to increase their energy demand during periods of PV generation, which is particularly advantageous in a REC. Also, this approach lowers overall energy demand by 12.6% and boosts self-sufficiency, and it also decreases electricity exports from the REC to the grid by 18.2% compared to a ToU tariff scheme. When using ToU tariffs, thermal storage can be used to achieve cost savings, but total Shared Energy decreases, as do self-sufficiency and self-consumption of the REC. The results indicate that in a REC with high variable renewable energy and decentralized control, consumers using TES and ToU tariffs with peak prices during high irradiance periods may not be beneficial for the grid compliance.
In conclusion, the coupling between DRL and thermal storage should be supported by more innovative pricing schemes for RECs and/or coordinated energy management, although it requires advanced communication and monitoring infrastructure.
This paper introduces a portable framework for developing, scaling and maintaining energy management and information systems (EMIS) applications using an ontology-based approach. Key contributions include an interoperable layer based on Brick schema, the formalization of application constraints pertaining metadata and data requirements, and a field demonstration. The framework allows for querying metadata models, fetching data, preprocessing, and analyzing data, thereby offering a modular and flexible workflow for application development. Its effectiveness is demonstrated through a case study involving the development and implementation of a data-driven anomaly detection tool for the photovoltaic systems installed at the Politecnico di Torino, Italy. During eight months of testing, the framework was used to tackle practical challenges including: (i) developing a machine learning-based anomaly detection pipeline, (ii) replacing data-driven models during operation, (iii) optimizing model deployment and retraining, (iv) handling critical changes in variable naming conventions and sensor availability (v) extending the pipeline from one system to additional ones.
Lab head

Department
- DENERG - Department of Energy
About Alfonso Capozzoli
- Alfonso Capozzoli graduated in Mechanical Engineering and obtained a PhD in Engineering of mechanical systems. He works as full professor at the Department of Energy of Politecnico di Torino. He teaches HVAC systems, building physics and energy management and automation in buildings. He is involved in various International Research Projects on building energy performance. He leads the Building Automation and Energy Data Analytics Lab (BAEDA Lab). Lab page: www.baeda.polito.it