Leandro Von Krannichfeldt
Web site: Web site: https://IMOS.epfl.ch
Fields of expertise
Energy forecasting, Signal Processing, Learning Algorithms
Biography
Leandro obtained his master's degree in Electrical Engineering and Information Technology from ETH Zürich. Before joining the Intelligent Maintenance and Operation Systems Laboratory, his research focused on energy demand and generation forecasting as well as collaborative learning. During the course of his Ph.D. study, he will explore hybrid digital twins for energy performance optimization.Research
Hybrid Digital Twins for Building Energy Systems
The operation of buildings is responsible for 30% of energy consumption and 26% of CO2emissions worldwide. Increasing digitalization and use of sensors hold great potential for
constructing digital twins for building operation optimizations. A digital twin is a continuously
monitored digital replica of the physical building, capable of forecasting future states and
suggesting control actions for more efficient operation. The foundation of a digital twin for
building energy optimization is based on the Building EnergyModel (BEM), a digital model used
for building energy analysis and prediction.
In building energy modeling, three major challenges related to data availability persist. First,
data scarcity is a common issue for newly built or newly equipped buildings with sensors, as
there may be insufficient data collected post-installation to develop a reliable model. Second,
after model development, newly incoming sensor data can alleviate data scarcity but poses
challenges for updating the model. Third, the unique sensor configuration in each building
complicates a unified exploitation of traditional building energy modeling, given that buildings
may not have the same number or types of sensors due to costs or equipment constraints.
This Ph.D. project aims to address these challenges by combining data-driven methodologies
with physics-based modeling into hybrid models to exploit their respective advantages.
Furthermore, we utilize knowledge from existing monitored buildings to benefit buildings
with low data availability. To tackle the data scarcity problem in a new building, we propose
leveraging a BEM created from a building with sufficient data. For this purpose, we use a hybrid
model that incorporates various physical constraints and focuses on bridging the gap in building
characteristics and operating condition between the buildings. To address the continuous
model update problem, we propose to transfer a hybrid model created from a building with
sufficient data to a target building experiencing data scarcity. Our focus here is on investigating
different hybrid models and adjusting them for continuous adaptation from incoming sensor
data streams. To address the unique sensor configuration challenge in buildings, we propose
constructing a generalizable model from a diverse and large dataset. We concentrate on
generating a dataset that covers variability in building characteristics, operating conditions, and
weather conditions, as well as investigating strategies to adapt the model to the target building.
The proposed methods will be evaluated using both simulated and real-world data from
residential and office apartments from EMPA, assessing model performance, interpretability,
and generalization ability.