My research focuses on the compositional safe learning and control of cyber-physical systems (CPS) with non-Markovian tasks, focusing on applications in power systems, particularly microgrids. Non-Markovian tasks, such as those expressed through temporal logic specifications or finite state automata, present significant challenges in control design due to their complex, often infinite, nature. Traditional approaches to Signal Temporal Logic (STL) tasks typically focus on simple, finite tasks, but in this work, we extend these methods to manage complex and industrially relevant temporal specifications, notably in the context of microgrid management. The core of this research is the development of a hierarchical multi-rate architecture for controlling microgrids, which consist of components such as batteries, supercapacitors, loads, and photovoltaic (PV) systems. The architecture is divided into two layers: a higher layer tasked with optimization, balancing cost minimization and safety constraints, and a lower layer that implements stabilizing controllers to ensure that the system states evolve as dictated by the higher layer. The number of control variables differs significantly between the two layers, introducing nonlinearities and complex interactions between discrete and continuous dynamics.
The thesis is funded by the French National Research Agency (ANR), supervised by Alessio IOVINE (L2S) and Adnane SAOUD (L2S).
