An Random Control Policy For Electricity Devices In Residential Demand Response Based On Online Learning 1

Residential demand response is of greater importance in the smart grid. While under the condition of dynamic energy price and uncertain energy request from the consumers, it is a particular problem to explicitly model this uncertainty and schedule the devices. In this paper we use markov model to depict this uncertainty, and formulate an optimization problem to minimize the average total cost in an infinite horizon. Because it's hard to get the transition probability information about the energy price and energy request, we derive a parameterized online learning algorithm to determine the power allocation of deferable electricity appliance and control the storage. A numerical example show that this algorithm is effective and can significantly reduce the long term cost.