PI: Ram Rajagopal; Co-PIs: June Flora, Hao Wang,
Oskar Triebe, Chad Zanocco, Chin-Woo Tan, Gonzague Henri
(Total) and Dimitra Ignatiadis (Total)
In recent years, utilities began deploying smart meters for residential customers unlocking a large amount of data. The first use case was billing automation, but with
the development of data science, cloud computing and cheaper data storage we can now develop new use cases and generate more value from this data.
In this project, Total and Stanford look at different data science techniques that could increase customer engagement. This project has two parts: algorithms and behavioral science. During this first year of collaboration, the team focused on developing the algorithm part of the project, the second year focusing more on the application and interactions with clients.
The following topics were investigated this year: monthly energy consumption forecasting, EV detection from AMI data, load disaggregation, activity detection, applying behavioral sciences to energy, federated clustering (privacy ML), and applying deep learning to time series.
Among the main results: the monthly bill forecasting task was completed, many update presentations were made to Total, an innovative alternative to standard load disaggregation looking at disaggregating activities rather than appliances was proposed, and a promising approach to bridge the gap between time series and deep
learning was developed. In order to maximize customer engagement, a review of behavioral techniques that could be applied to the energy sector was completed.