With the increasing penetration of variable renewable energy

flexible resources are highly needed to hedge the growing uncertainty

and variability in the power system. Demand response has served as a cost-effective type of flexible resource in recent years. In order to balance the uncertainty of the system

it is crucial to assess how much flexibility demand response programs can provide. Thus

forecasting demand response potential is important for the operation of the bulk system. This paper proposes a modeling approach that can characterize the multi-timescale flexibility of demand response so that not only the power potential but also temporal-coupling characteristics can be considered. Furthermore

a day-ahead demand response potential forecasting method is proposed using deep convolutional generative adversarial networks. The proposed forecasting method is tested using data from 170 users in Pecan Street Dataport. The results show that the proposed method can forecast the multi-timescale flexibility of demand response with high accuracy.