Reinforcement Learning for the Management of an Electrical Vehicle Fleet in A Distribution Grid

With the increasing share of Electrical Vehicles (EVs) in the transportation sector, the impacts on the power distribution grid are also non-neglectable. The increasing line loadings and voltage derivations due to the increasing load are the most significant impacts on the distribution grid. To manage EV fleet charging, in recent years, Deep Reinforcement Learning (DRL) has drawn more attention due to its inherent uncertainty mitigation ability. However, the DRL methods falls short when dealing with complex time series-related problems. In this paper, we proposed DRL-based EV fleet charging management in a distribution grid to mitigate the impacts of the EV fleet on the distribution grid, i.e., line loading and bus voltage derivations. The DRL agent is built with Long Short-Term Memory (LSTM) to enhance time series processing. We investigated two structures of LSTM agents and the numbers of LSTM layers. The results show that the LSTM agent is efficient to reduce EV impacts on grid, that reducing 47% current overloading and 33% voltage out-of-limit while deliver 87% energy to EV fleet.