A Market Oriented, Reinforcement Learning Based Approach for Electric Vehicles Integration in Smart Micro Grids

Abstract

In an independent self-sustained micro grid (MG) with limited energy resources, plugged-in electric vehicles (EV) must compete for available excess power supply or demand, modeled as a random variable. This paper proposes a distributed machine learning algorithm based on a Markov decision process (MDP) and non-cooperative game theory, that maximizes the EV’s profit under uncertainty of future MG supply/demand states, while satisfying specific battery constraints imposed by the EV owner. Performance evaluation of the proposed algorithm shows that even with no a priori knowledge of future MG supply/demand states, it achieves average profits of only 43% less than the global optimal profit. Results also show that using a cooperative version of the algorithm leads to a 12% increase in average profits.