ML-based Energy Modeling Could Optimize Sustainable Designs

The U.S. building sector accounts for almost 40% of annual greenhouse gas emissions.  The emissions from the building sector come from building energy consumption and inefficiency, the materials and equipment used in construction, and operational emissions.

 

The building sector contributes to almost 40 percent of greenhouse gas emissions annually. Image used courtesy of Pixabay

 

Designers are encouraged to complete energy modeling to project and better understand the designs’ expected energy consumption. 

 

Energy Modeling

Energy modeling can assist in finding the most effective energy-efficient design strategies for projects, such as optimizing the building orientation, selecting energy-efficient systems and materials, and incorporating renewable energy technologies. Simulating energy performance can also evaluate the life-cycle cost of a building, including the long-term energy costs.

Designers use several energy modeling software programs; however, these tools demand significant manual input of comprehensive parameters encompassing design and operational variables, including various building characteristics (materials for walls, building envelope, and windows) and operational parameters (expected climate and location temperatures).

 

Machine Learning Techniques

Researchers from Florida Tech are conducting a study to investigate whether machine learning can aid in reducing the environmental impact caused by buildings. 

Published in the journal Energies, the study employs a novel approach to building energy modeling (BEM) and BEM optimization. The study authors, Hamidreza Najafi, an associate professor of mechanical engineering, and Benjamin Kubwimana, a mechanical engineering graduate student, are developing a Python-based software script to facilitate the automated entry of data into EnergyPlus, a physics-based building energy simulation tool. 

 

Emerson used EnergyPlus and Python EMS to optimize refrigeration and HVAC, delivering building-level energy savings and load flexibility. Image used courtesy of National Renewable Energy Laboratory (NREL)

 

By introducing a range of variables as inputs through this Python script, a broad spectrum of multiple parameters can be accounted for, resulting in the generation of extensive datasets, which can be utilized to create a surrogate energy simulation model. 

The study trains machine learning algorithms, specifically artificial neural networks, from these datasets. The surrogate model is subjected to two optimization methods—genetic algorithm and Bayesian optimization—to attain the optimal building design. This approach can be customized to incorporate diverse design or operational parameters.

This system could greatly affect the efficiency and accuracy of energy modeling.

 

Lifecycle Applications

A larger initiative as part of the study was to improve BEM and broaden its applications. Improved BEMS can serve as digital replicas of buildings, offering benefits to owners and developers, during the construction phase and throughout the building’s life cycle.

The researchers have also developed a way to automatically gather data from building sensors and input it into the computer models. This allows the digital replica of the building to be continually updated to reflect its current operating state. This could aid the building owner in estimating energy consumption based on changes in operational parameters. It would enable proper budget planning for energy costs and project energy use while decreasing the associated emissions.