Distributed Ensembles of Reinforcement Learning Agents for Electricity Control

Abstract-- Deep Reinforcement Learning (or just "RL") is In this paper, we aim to answer them. Then, we aspects: intermittent nature of renewable energy, variations evaluate the computing cost of the building phase and the in demand, low storage abilities, [1] [2] significant room inference phase running on modern computing nodes. Deep This paper first demonstrates experimental evidence that Reinforcement Learning has shown great success in scaling homogeneous ensembles with averaging as a combination up model-free reinforcement learning algorithms to the rule are more performant and stabler than one individual RL challenging Markov Decision Processes [4] [5] and is a agent and other ensemble procedures. Second, we perform promising method to solve issues of electricity control [6]. Finally, due to the simplicity To alleviate this, we analyze and propose an ensemble of of the proposed procedure and the stabilization effects, our deep reinforcement learning agent procedures and discuss its experiments are easily reproducible.