Post-Doctoral Research Visit F/M Real-time learning methods for net-consumption control on a power grid

The Inria Saclay-Île-de-France Research Centre was established in 2008. It has developed as part of the Saclay site in partnership with Paris-Saclay University and with the Institut Polytechnique de Paris .

The centre has 40 project teams , 32 of which operate jointly with Paris-Saclay University and the Institut Polytechnique de Paris; Its activities occupy over 600 people, scientists and research and innovation support staff, including 44 different nationalities.

This proposal is part of the Inria - EDF challenge entitled 'Managing tomorrow's power systems'.

More specifically, this proposal is supported by
- the Inria Celeste (https://team.inria.fr/celeste/) and Thoth (https://team.inria.fr/thoth/) project-teams
- the Consumption Forecasting (R39) and Upstream Optimization (R36) teams of EDF R&D's OSIRIS department.

The proposed supervision team is made up of
- on the Inria side: Pierre Gaillard (Thoth), Hadrien Hendrikx (Thoth), Gilles Stoltz (Celeste)
- on the EDF side: Nadia Oudjane (R36) and Margaux Brégère (R39)

The position will be based in the Celeste team in Orsay, with a weekly split between Orsay and EDF R&D (center located in Palaiseau). Regular travel to Grenoble, to the Thoth team, is also planned for this position (travel expenses covered within the limits of the applicable regulations).

Proposed research subject 

New communication tools between electricity suppliers and their customers enable to control consumption, either directly by controlling certain uses (water heaters, swimming pool pumps, electric vehicles [EVs]), or indirectly by sending incentive signals (price, off-peak hours). This enables to adjust consumption to the production of renewable energies. The aim of this research proposal is to extend existing work to real-time load management, to get as close as possible to a target consumption whose accuracy would evolve over time, as well as adapting in real time to the constraints and hazards of usage.

A number of studies have already been carried out on the management of electricity consumption through the servo-control of certain equipments (optimization of swimming pool pump operation in Florida, control of residential equipment, smart charging of electric vehicles, etc.); see, for example:
- S.P. Meyn, P. Barooah, A. Bušić, Y. Chen, and J. Ehren. Ancillary service to the grid using intelligent deferrable loads. IEEE Transactions on Automatic Control, 60(11):2847-2862, 2015
- R. D'Hulst, W. Labeeuw, B. Beusen, S. Claessens, G. Deconinck, and K. Vanthournout. Demand response flexibility and flexibility potential of residential smart appliances: Experiences from large pilot test in Belgium. Applied Energy, 155:79-90, 2015
- M. Zweistra, S. Janssen, and F. Geerts. Large scale smart charging of electric vehicles in practice. Energies, 13(2):298, 2020

The following description is based on the following article [A], written by members of the supervision team:
- [A] B.M. Moreno, M. Brégère, P. Gaillard, and N. Oudjane. Reimagining demand-side management with mean field learning. arXiv preprint arXiv:2302.08190, 2023

Article [A] proposes an iterative learning algorithm to optimize load steering decisions to be sent to a set of consumers over time to satisfy a global target (typically net electricity consumption). Control is achieved by monitoring various household appliances---in this case, water heaters. However, [A] relies on a priori optimization, based on past or simulated data, assuming that the target consumption is deterministic and known precisely at the time the decision is made. The method developed by [A] does not allow decisions to be adapted in real time according to observed usage (e.g., the presence or absence of electric vehicles), observed contingencies (e.g., meteorological), or changes in the target. In practice, the target to be reached is not deterministic, but is obtained from forecasts of renewable consumption/productions. This target is therefore random, and its accuracy varies over time. The work of [A] needs to be adapted to this context: how can a probabilistic target (probably increasingly accurate) be incorporated into algorithms in real time?

In this proposal, we propose to extend the work of [A] to real-time optimization for load control (typically electric vehicle charging) of a set of consumers on a network to take into account the practical issues detailed above. The probabilistic forecasts will be derived from results obtained during Guillaume Principato's PhD thesis, which focuses on hierarchical conformal forecasting. A second aspect of this proposal will be to introduce tools for evaluating probabilistic forecasts by exploiting the gains induced by real-time optimization.

We also expect other avenues of research to emerge, depending on the specific problems of the uses considered: wear and tear of equipments (water heaters, EV6 batteries, etc.) in the event of too-frequent changes in charging/discharging, non-stationary aspects of behavior (rapid evolution of electrical uses), hierarchical aspects of the electrical network and the target.

The research mission includes the production of both theoretical and practical contributions, to be enhanced by:
- publications and presentations in machine learning or optimization conferences or journals,
- creation of Python or R packages,
- writing internal reports if EDF proprietary data is used.

A more general involvement in the Inria - EDF challenge is also expected, such as participation in the monthly seminar, interaction with other sub-projects of the challenge, and providing input for activity reports.

Candidates must hold a Ph.D. in Applied Mathematics, Computer Science or a closely related discipline. Candidates must also show evidence of research productivity (e.g., conference proceedings or journal articles) at the highest level.

We prefer candidates who have a strong mathematical background (on optimization and/or machine learning) and in general are keen on using mathematics to model real problems and get insights.

Thes candidates should also have
- good programming skills, either with R or Python,
- good communication and reporting skills, with proficiency in English,
- an interest in collaborative work.

• Subsidized meals
• Partial reimbursement of public transport costs
• Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
• Possibility of teleworking and flexible organization of working hours
• Professional equipment available (videoconferencing, loan of computer equipment, etc.)
• Social, cultural and sports events and activities
• Access to vocational training
• Social security coverage

Gross Salary : 2788 € per month