Concept for hierarchical load management in MV-LV distribution networks using reinforcement learning with master-slave structure

Grid integration of electric vehicles (EV) is a big challenge for distribution grids, since simultaneous charging processes of large number of EVs could result in grid congestion, such as overloads in transformers and lines. Considering that EVs are connected to power systems at different voltage levels, a cooperated use of EV flexibility across multiple voltage levels can be beneficial for relieving grid congestion. This paper proposes a hierarchical load management system with a master-slave structure, which relies on local measurement data and is performed in a distributed manner. Reinforcement learning (RL) can be used to acquire control polices for each control unit. The framework and the information flow of the whole system are introduced and analyzed. In addition, the algorithm based on RL is described in detail in this concept paper.