Regression markets and energy forecasting applications

Pierre Pinson (Imperial College London)

Friday 21st April, 2023 15:00-16:00 Maths 311B

Abstract

The operation of energy systems heavily relies on data, where most agents would benefit from also accommodating data (or more generally information) for other agents. There does not exist, however, a general framework that would allow incentivizing information sharing, with the general objective of improving energy system operation in a liberalized market environment. So far, data has largely been taken for granted as a free and highly accessible commodity in energy systems operations, which is in glaring contrast to the growing concern over privacy both on small individual energy user levels, and on large corporate or even national levels. We hence propose to explore designs for data marketplaces that would be relevant for energy systems. As a special case, emphasis is placed on data markets linked to specific analytics tasks e.g. regression as a support to forecasting (may be least-squares or quantile regression for instance). Our proposal specifically focuses on yielding the right market properties, e.g., to incentivize data sellers to provide high-quality data while being given the freedom to set their individual return threshold based on privacy. Meanwhile, the data buyer balances the trade-off between the payment to the data sellers and their own gain from the additional data. Those proposals are made within both batch and online learning setups, to generally accommodate different types of analytics tasks within energy system operations. Email wei.zhang.2@glasgow.ac.uk for Zoom link.

Add to your calendar

Download event information as iCalendar file (only this event)