Abstract

The AI4DG German-French joint research project focuses on AI on the edge for a secure and autonomous distribution grid control with a high share of renewable energies. It is a highly international competitive project awarded under the ANR/BMBF collaboration.

Coordinators of the project

• Nouredine Hadj-Said: Head of the MIAI chair "Artificial intelligence for Smart Grids", member of the coordination team.

Description

This international collaboration involves two academic partners (specialized in electrical engineering and applied computer science and Artificial Intelligence in France and Germany - Univ. Grenoble Alpes and Univ. Bielefeld) associated with two industrial partners (ATOS worldgrid and Stadtwerke Versmold) in order to work on improving the management of flexible resources on low voltage networks using Artificial Intelligence techniques.

As such, the objective of the AI4DG project is to study and develop an AI platform whose role would be to ensure secure and autonomously control distribution grids presenting a high share of renewable with the help of distributed storage units. For such a system, AI can estimate the current state of the grid, predict power generation from renewable generation sources, and ultimately optimize system services by leveraging battery charging and discharging.

A complex system, in which the components of a future electrical network are connected to each other, is prone to failures, especially if the control of the components is carried out in a strictly centralized manner. To ensure safe and reliable operation of the network, communication needs to be secure and to ensure that AI operates securely, a decentralized approach with AI at the edge is therefore considered: for example, if an AI on a distributed node fails, another unit can take control. For this approach to AI at the edge, Cognitive-Edge-Computing is considered to allow effective control, reducing the necessary hardware resources (CPU, data storage, communication exchanges) and ensuring data security and confidentiality.

The AI methods studied and developed and the Cognitive Edge Computing architecture will be validated in a network simulation laboratory, implemented on the management platform proposed by the industrial partner and evaluated in field experiments on the electrical networks of the distribution grid operator.

The case study currently considered represents a real low-voltage distribution network on which an optimization will manage local production and consumption (with storage) in order to meet several operational criteria of the network operator, as well as the individual welfare of the energy end-users. The optimization will be based on a consumption prediction algorithm. A decentralization of intelligence (including decision-making) will be studied in support of the centralized control operated by the network manager, who will exchange data and models with the decentralized control units initially, then only models at the finalization of the work.