Trainee Project
Title:
Optimization of neural network training for battery state-of-charge estimation
Dates:
2024/04/02 - 2024/09/30
Supervisor(s):
Other supervisor(s):
CLAUSEL Marianne (marianne.clausel@univ-lorraine.fr)
, Daniel Monier-Reyes
Description:
Background: Worldwide economic demand for electrochemical batteries is on the increase. This is mainly due to the emergence of hybrid and electric
vehicles (Hybrid-Electric Vehicle, Plug-in Hybrid Electric Vehicle and Battery-Electric Vehicle) on the one hand, and the energy storage market linked to
renewable energies and power grid management on the other.
SAFT is particularly active in this context, as a pioneer in the deployment of lithium batteries. SAFT produces lithium-ion batteries in Poitiers, Nersac and
Bordeaux. This internship, financed by SAFT, will take place at CRAN in Vandoeuvre-lès-Nancy.
Subject description: The aim is to study the problem of estimating the state of charge of industrial batteries using neural networks. The central issue will be
the search for optimal architectures, either automatically (Neural Architecture Search, AutoML) or more conventionally. The work will be completed by a
study of the impact of the choice of architecture on estimation uncertainties.
Study plan: The first step will be to build on the work already carried out at SAFT, where seq2seq neural network models have already been developed to
solve the problem.
Secondly, the trainee will investigate various up-to-date architecture search methods (Neural Architecture Search, Auto ML).
The final stage will involve modifying the various architectures to add constraints including control of estimation uncertainty.
References :
[1] Elsken, Thomas, Jan Hendrik Metzen, and Frank Hutter. "Neural architecture search: A survey." The Journal of Machine Learning Research 20.1 (2019):
1997-2017.
[2] Kabir HD, Khosravi A, Hosen MA, Nahavandi S. Neural network-based uncertainty quantification: A survey of methodologies and applications. IEEE
access. 2018 Jun 4;6:36218-34.
vehicles (Hybrid-Electric Vehicle, Plug-in Hybrid Electric Vehicle and Battery-Electric Vehicle) on the one hand, and the energy storage market linked to
renewable energies and power grid management on the other.
SAFT is particularly active in this context, as a pioneer in the deployment of lithium batteries. SAFT produces lithium-ion batteries in Poitiers, Nersac and
Bordeaux. This internship, financed by SAFT, will take place at CRAN in Vandoeuvre-lès-Nancy.
Subject description: The aim is to study the problem of estimating the state of charge of industrial batteries using neural networks. The central issue will be
the search for optimal architectures, either automatically (Neural Architecture Search, AutoML) or more conventionally. The work will be completed by a
study of the impact of the choice of architecture on estimation uncertainties.
Study plan: The first step will be to build on the work already carried out at SAFT, where seq2seq neural network models have already been developed to
solve the problem.
Secondly, the trainee will investigate various up-to-date architecture search methods (Neural Architecture Search, Auto ML).
The final stage will involve modifying the various architectures to add constraints including control of estimation uncertainty.
References :
[1] Elsken, Thomas, Jan Hendrik Metzen, and Frank Hutter. "Neural architecture search: A survey." The Journal of Machine Learning Research 20.1 (2019):
1997-2017.
[2] Kabir HD, Khosravi A, Hosen MA, Nahavandi S. Neural network-based uncertainty quantification: A survey of methodologies and applications. IEEE
access. 2018 Jun 4;6:36218-34.
Keywords:
Machine learning, batteries, python, optimization, neural network architectures
Conditions:
The internship will take place at CRAN, UMR CNRS 7039, Vandoeuvre-lès-Nancy (54).
Department(s):
| Biology, Signals and Systems in Cancer and Neuroscience |
Funds:
Funded by SAFT