2022/3/28 10:31:56 阅读：331 发布者：chichi77
PhD fellowship - Combination of physics-based models and Machine-learning for multi-energy systems modeling
Summary:
In the frame of the energy transition, Multi-Energy Systems (MES) offer the opportunity to increase global efficiency and decrease the environmental footprint of human activities. Developing fast and precise models for such systems is a topical research subject. These models must be realistic and adapted to real-time applications to enable planning and optimal control of energy systems including production, distribution, and storage systems. Thus, these models must have a reasonable computational cost and should not require a great amount of input data for their calibration. Hybrid simulations could be an interesting approach that is still insufficiently studied in the scientific literature. This approach encompasses both “Simulation Assisted Machine-Learning (SAML)” and “Machine-Learning Assisted Simulation (MLAS)”. The goal of this Ph.D. is to study the contributions of hybrid modeling to the design and optimal control of MES.
Research questions:
The main goal of the Ph.D. work is to compare different hybrid modeling strategies in terms of representativeness, computational performance, and data requirements. Models of MES will be investigated at the component and/or the system’s levels. Different degrees of hybridization will be tested from pure physics-based models to pure machine-learning models. To achieve this main goal, four intermediary objectives should be reached:
A methodology for splitting the models into relevant sub-models or elementary blocs will be developed. Sub-models where ML modeling is appropriate (trained by measured or simulated data) and those for which physics-based solutions are more suitable, will be identified. This identification will be based on the predictability and the complexity of the sub-model. This first goal will lead to a validated methodology to decide which hybrid modeling architecture should be adopted for each selected simulation case.
The comparison of the different modeling strategies will rely on the assessment of KPIs that measure their precision, computational cost, and data requirement. The outcome of this stage will be the identification of the best architectures and degrees of hybridization for modeling MES.
Considering the dynamic behavior of MES which have different response times will enable assessing the relevance of using hybrid modeling approaches in such conditions.
Developing dynamic models will enable studying MES with the objective of developing model-based control strategies and assessing the benefits of these strategies along with hybrid modeling in terms of energy efficiency and greenhouse gases emission.
Simulations performed in this Ph.D. work will rely on the Gemellus platform which is developed at IMT Atlantique and dedicated to modeling and optimization of multi-carrier energy systems. This work will support and extend the development of this platform.
Plan of the study:
1.Following the objectives stated previously, the first step of the study will consist of selecting two case studies: a multi-carrier energy conversion technology (one component) and a multi-carrier energy system (combination of components). These case studies will be used to assess the different modeling strategies.
2.These case studies will be first modeled by using a purely physics-based approach (Mod.1) which is the paradigm used by Gemellus. The results obtained will serve as a reference that will
Mod.2: based on the MLAS paradigm where ML models are used for the identification of parameters used by physics-based models give a first idea about the nature of input data required for the simulation process (quantity, quality, predictability, and variability). Then, three other approaches will be developed and tested:
Mod.3: based on SAML paradigm where physics laws are used to guide the training process of ML models.
Mod.3: totally ML based models that will be used both for physical modeling and parameters identification.
3.For relevant strategies, the parameters that should be substituted by ML models will be selected based on the analysis of the input data required for the simulation process as detailed previously. In addition, different architectures of hybrid models will be tested and studied.
The comparison of the different modeling strategies using a rigorous methodology will constitute the major scientific outcome of this Ph.D. work. On the other hand, demonstrating the benefits of using hybrid models in model-based control of MES will constitute the major operational outcome.
Background and skills:
The applicant should have either a background in Mechanical or Energy Engineering with proven experience in developing Machine Learning models or, a background in Machine Learning with proven experience in modeling physical systems (energy).
The applicant will demonstrate curiosity, analytical thinking and capacity to propose ideas.
Excellent mastery of scientific programming and algorithmic will be a very important selection criterion.
Good knowledge of one or more Machine learning frameworks is strongly recommended.
Application deadline: April 30th, 2022
Starting date: September/October 2022
Contacts:
Bruno Lacarrière, IMT Atlantique, Department of Energy Systems and Environment
Email: bruno.lacarriere@imt-atlantique.fr
Mohamed Tahar Mabrouk, IMT Atlantique, Department of Energy Systems and Environment
Email: mohamed-tahar.mabrouk@imt-atlantique.fr
Title
PhD fellowship - Combination of physics-based models and Machine-learning for multi-energy systems modeling
Employer
IMT Atlantique
Location
4, rue Alfred Kastler - La Chantrerie Brest, France
Application deadline
2022-04-30
Job type
PhD
Field
Computational Engineering, Computational Physics, Computing in Mathematics, Natural Science, Engineering and Medicine, Energy Technology, Engineering Physics and 3 more
如有侵权，请联系本站删除！
