IEEE Transactions on Smart Grid · 2020 · 203 citations · 27 references
ComputingComputer ScienceArtificial IntelligenceMachine LearningSchedulingDeep LearningDemand ResponseDemand-side ManagementPolicy LearningReal-time OptimizationReinforcement LearningAppliance SchedulingControl ActionsDeep Reinforcement LearningDemand Response OptimizationDemand Response SchedulingModel-free Reinforcement LearningReinforcement Learning For SchedulingTrust Region Policy Optimization
This paper presents a real-time demand response (DR) strategy for optimal scheduling of home appliances. The uncertainty of the resident's behavior, real-time electricity price, and outdoor temperature is considered. An efficient DR scheduling algorithm based on deep reinforcement learning (DRL) is proposed. Unlike traditional model-based approaches, the proposed approach is model-free and does not need to know the distribution of the uncertainty. Besides, unlike conventional RL-based methods, the proposed approach can handle both discrete and continuous actions to jointly optimize the schedules of different types of appliances. In the proposed approach, an approximate optimal policy based on neural network is designed to learn the optimal DR scheduling strategy. The neural network based policy can directly learn from high-dimensional sensory data of the appliance states, real-time electricity price, and outdoor temperature. A policy search algorithm based upon trust region policy optimization (TRPO) is used to train the neural network. The effectiveness of our proposed approach is validated by simulation studies where the real-world electricity price and outdoor temperature are used.
27
Human-level control through deep reinforcement learning
Volodymyr Mnih, Koray Kavukcuoglu, David Silver et al. · Nature · 2015 · 30K citations
Mastering the game of Go with deep neural networks and tree search
David Silver, Aja Huang, Chris J. Maddison et al. · Nature · 2016 · 15.8K citations
Trust Region Policy Optimization
John Schulman, Sergey Levine, Philipp Moritz et al. · arXiv (Cornell University) · 2015 · 3.1K citations · Full text