Feedback Experience : Inria Innovation Lab METIS
Artelys is an SME specialising in optimisation solutions, statistics and decision support. The company has become a leader in the field in France, and is expanding in the international market. Artelys provides solutions and services ranging from strategic optimisation to developing operational software solutions.
From 2010 to 2014, within the Inria Innovation Lab Metis, Artelys and Inria designed and developed algorithms to optimise large-scale energy systems in which future supply and demand are uncertain and in which energy storage plays a key role.
In particular, they developed the DVS algorithm (Direct Value Search). This is used to solve the problem of how to optimise management of electricity storage systems, including hydraulic systems, and has many advantages over conventional algorithms used in this sector, such as MPC (Model Predictive Control) and SDDP (Stochastic Dual Programming): approximate results are immediately accessible, and they do not rely on assumptions regarding either cost convexity or random processes.
Integrating this algorithm into the Artelys Crystal operational solution has been successfully tested. Artelys and Inria are pursuing their collaboration in the field of energy system optimisation with the POST project (designing large-scale electric power systems).
Marc Schoenauer, directeur de recherche Inria et responsable de l'équipe TAU (ex TAO)
The experience of TAO within the framework of the Métis Innovation Lab has been most rewarding: not only for the ensuing contractual impacts (an FP7 project and two projects backed by the Investments for the Future programme undertaken by ADEME - POST (French Environment and Energy Management Agency - Transcontinental Supergrid Optimisation Platform) and the NEXT project) but, above all, for the acculturation of the team to the energy management issues it has brought. We have discovered, 'from the inside', the real problems faced by industry, to all appearances quite similar to academic issues, but with subtle differences that lead to real research subjects - such as, for example, the consideration of non-stochastic uncertainties in investment optimisation problems on large electrical networks. Conversely, we are particularly proud of having contributed to the acculturation of the industrial world to cutting-edge optimisation and machine learning techniques, for example concerning the systematic use of cross-validation for more robust results faced with the unknown that is the future.