Reinforcement Learning applied to Smart Charging algorithm optimization

Fábio Miguel Ribeiro Ramos

Key information

Authors:

Fábio Miguel Ribeiro Ramos (Fábio Miguel Ribeiro Ramos)

Supervisors:

Sérgio Luís Proença Duarte Guerreiro (Sérgio Luís Proença Duarte Guerreiro); José Manuel Granate Marques (José Manuel Granate Marques)

Published in

November 24, 2022

Abstract

The world is currently experiencing an energy transition and most distribution grids are reaching maximum capacity, giving new importance to the intelligent management of the scarce resource that is energy. Electric Vehicles (EVs), a technology that is gaining increasing attention in the past years, are putting even more pressure on these distribution grids. With this increase in demand, the problem of managing the charging process of multiple EVs with a very low limit of power available at a certain instant stands as one of the main obstacles to the evolution of electric mobility infrastructures. In this work, a new system based on Reinforcement Learning (RL) is presented to tackle this problem to optimize the smart charging of Electric Vehicles in a residential Scenario. In other words, the system pretends to enable the charging of several EVs in a timely manner by allocating power limits to the charging stations through protocols applied worldwide like OCPP. To evaluate the agent, the interaction between it and the various scenarios was registered and analyzed to understand the efficiency of the system. Studies have shown promising results when applying RL methods as energy management strategies, saving up small quantities of energy and using energy more efficiently from Photovoltaic systems or batteries, hence reducing the chances of overloading the grid. This thesis creates an innovative solution for EV charging scheduling optimization, with a focus on Battery Electric Vehicles (BEVs), and to facilitate the energy transition that is happening, by creating and improving the electrical infrastructures.