Scientific direction Development of key enabling technologies
Transfer of knowledge to industry

PhD : selection by topics

Technological challenges >> Smart Energy grids
2 proposition(s).

See all positions [+]

Pre-sizing method for multi-source hybrid systems includin the energy management strategy

Département de l'Electricité et de l'Hydrogène pour les Transports (LITEN)

Laboratoire Architecture Electrique et Hybridation



Smart Energy grids (.pdf)

Embedded hybrid energy source systems are popular for the energy transition but are also complex to size: there is a strong dependence between the design of energy sources (battery, fuel cell, wind turbine, internal combustion engine, etc.) and the choice and configuration of the control law in real time (hybridization rate, real time strategy). The larger the number of sources, the more "complex" the problem to be solved is: the number of decision variables and the couplings between them become too important for traditional optimization methods. Pre-sizing such systems requires a new method of optimization. According to the state of the art of research on the pre-sizing by optimization of hybrid systems including the control law, the method will probably have to be based on a hybridization of different optimization techniques (branch and bound, stochastic, multi-level, quadratic, etc.). The objective of this hybridization would be to correctly manage the interdependencies of the energy sources with the control law during the pre-sizing phase while guaranteeing an optimal system with a reasonable calculation cost. The method proposed in the thesis could be tested, analyzed and improved on different applications such as the Energy Observer.

Download the offer (.zip)

Thermal networks fault detection and localization on a citywide scale: an approach combining artificial intelligence and physical simulation

Département Thermique Biomasse et Hydrogène (LITEN)

Laboratoire des Systèmes Energétiques et Démonstrateurs Territoriaux



Smart Energy grids (.pdf)

Detection and localization of faults in thermal networks on a citywide scale: an approach combining artificial intelligence and physical simulation Over time, thermal networks (heating and cooling networks) age differently. Damages to underground pipelines, which can be invisible and not repaired, endanger not only the financial equilibrium of the network operator, but also the quality of heat supply to users, especially in winter periods. Thus, detecting hydraulic leaks and other anomalies remains one of the network operator's top priorities, as it requires a significant share of investment related to civil engineering works for locating and repairing leaks. The objective of this thesis is to implement an innovative diagnostic approach for the detection and localization of anomalies on the pipes of a heating network by exploiting network measurement data (data at the substation level and data from detectors positioned in the gutters) and the capabilities that artificial intelligence offers. The main objectives of this thesis are the following: - Generate database of normal and abnormal network operating modes using numerical simulation. This will be based on existing numerical models in the host laboratory. - Use numerical models for data validation and reconciliation to improve the quality of measurement data from a real network. - Identify algorithms for the detection and the localisation of anomalies based on machine learning algorithms (AI), data generated by simulation and measurement data from a real network. - Validate the performance of the algorithms for detecting anomalies on various simulated operating scenarios and then in real-life situations. - Implement a decision support tool for the detection and localization of anomalies for thermal networks One of the originalities of this work lies in the complementarity between the use of AI methods (supervised and unsupervised learning and classification) and detailed thermo-hydraulic models of networks. The latter should make it possible to compensate for two of the main pitfalls currently encountered by the use of AI methods: the strong dependence on the quality of measurement data, and the need to have very large databases with a high number of anomaly occurrences.

Download the offer (.zip)

See all positions