Title: Charging Electric Vehicles with MDPs and GPs
Abstract: Electric vehicle sales are rising, which not only has ecological benefits, but also brings about some new challenges. Assuming dynamic prices for end users, we try to optimize the charging process of such vehicles with the help of a smart charging agent. At the beginning of each charging process our agent refines its knowledge about the user’s preferences, predicts energy prices using Gaussian processes, and then solves a Markov decision problem to define a charging policy. A number of experiments show that our agent is more efficient and economical than our simpler comparison charging agents. Additionally, our agent reacts to price peaks by lowering its demand, which would help to reduce local energy consumption during peak times.
