ADRENALIN

Project description

Buildings represent a high share of peak electricity demand. Commercial and institutional buildings often already have equipment and control capability in place for delivering flexible demand into the grid. Consequently, buildings offer the potential to be one of the lowest cost opportunities for providing the flexible-demand needed to support increasing levels of variable renewable energy resources in electricity grids.

Unfortunately, it has proven difficult to activate and scale this latent flexible-demand opportunity. While digitalization is thought to be a key enabler, many of the existing digital platform solutions stop ‘at the building gate’ and have consequently proved ineffective at motivating adoption. To unlock this opportunity, much deeper connectivity is required into building equipment systems and much greater focus is required on digitalization use-cases, taken from the perspective of the building owner.

In short, new data-driven software services are needed, harnessing the power of machine learning, to deliver smart-grid-enabled buildings at scale.

The main objective of this project is to establish and support an ecosystem of research and industry innovators by organizing a series of machine learning competitions for data-driven smart building applications, aimed at higher-education students. This will be in collaboration with the IEA EBC Annex 81 "Data-driven Smart Buildings", in which most of the consortium's partners participate. This will be similar to the ASHRAE Great Energy Predictor III competition that was held in late 2019, which focused on the prediction/baselining of energy consumption in thousands of buildings worldwide, and will be structured into three phases:

1. The first phase will consist in the development of a large library of open data sets from built environment applications: a data sandbox. It will be required that such data sets have a certain level of documentation to be considered for the library
2. The second phase will be the launch of a Kaggle-based competition focused on various data-driven applications related to different topics
3. The third phase will involve implementation of highly-commended solutions as “Applications” on a commercially independent data platform, as a means of accelerating the pathway to industry adoption

A wide range of uses cases and competition topics are possible. Some examples that could be addressed are:

• Automated decoding of point labels into semantic building models
• Load disaggregation from main energy metering data
• Inferring building temperatures and indoor air quality from energy metering data
• Model Predictive Control applications

Beyond theoretical innovation, the practical implementation – especially in the highly distributed and fractalized building stock – is a tremendous challenge to create added value out of data-driven services. Therefore, the competition will aim to reward the practical applicability of data-driven solutions in real life projects.

Project summary