Secure, Reliable, and Predictable Smart Grid (smartgrid)

A project focusing on the security, reliability, and predictability of the electricity grid in Luxembourg for optimization and robustness purposes.

The project at a glance

Start date:

15 Mar 2021
Duration in months:

24
Funding:

CREOS S.A.
Principal Investigator(s):

Yves Le Traon

About

Since 2013, the SerVal research group at SnT has been actively collaborating with Creos S.A. on its smart grid activities, a partnership that recently entered its third iteration.

Indeed, the fast-paced deployment of smart meters all across Luxembourg enabled a whole new set of applications as well as raised challenges, ranging from daily operation optimization to load (and overload) forecasting.

The first project iteration (2013-2017) largely focused on the modelling of the grid, its components, and the integration of consumption data to enable near real-time analysis. This initial project led directly to the creation of an SnT spin-off named Datathings, that is still working on the topic alongside Creos.

Following this first successful collaboration, a second project (2017-2020) to explore grid robustness was started. During this iteration, the topics of noise detection in smart meter communication, fuse state detection, and grid optimization to prevent cable overloading were explored leading to several publications and prototypes.

Leaning on these outputs, Creos renewed the collaboration for a third iteration focusing this time on the security, reliability, and predictability of the grid. This current project is divided into two parts:

Subproject 1: Adaptive Big Data Prediction Model
Previous collaborations led to the definition of data models to predict customers consumption and the consequent cable load. However, it may happen that the assumptions supporting the model become invalid over time (e.g. customers move, installation of solar panels, purchase of an electrical car, etc.), leading to increased rates of incorrect predictions. These errors may, in turn, infect any decision model built on top of the data model. Thus, the main goal of this axis is to investigate context-aware models that:
(1) provide the level of confidence associated with the prediction they make;
(2) implement strategies to adapt themselves to changing assumptions (e.g. retraining, online learning, model catalog…).

The research will be carried out within the full extent of Luxembourg’s smart grid, which makes it necessary to deal with a large amount of data.

Subproject 2: What-If Analysis for Robust Smart Grid
In addition to reacting to incidents (e.g. overloading) by changing the grid configuration, the goal is to minimize the risk for the new configuration to run into undesired states. Accordingly, this second axis aims at designing methods to evaluate the adequacy of the new configuration over time, considering the inherent uncertainty of the future states of the grid, and the external events that affect it. A specific part of such evaluation would stress-test the grid against rare but impactful events (cable break, populated events, etc.) to assess its robustness. This necessitates:
(1) techniques to simulate rare events (mutating load values and grid topology) and
(2) to incorporate the robustness analysis in the overall decision model for suggesting grid configuration.