A Novel Event Data Augment Method for Characterizing Price Responsive Demand

Price responsive demand is recognized as one of potential resources for shaving peak load demand and prompting renewable power accommodation in power system dispatch. To characterize the price responsive demand, datadriven characteristic assessment, which needs massive demand response (DR) event data, is common employed. The event sample contains key information, such as response load, price signals and seasons. It is difficult to obtain large amount of realistic event data in practice. In this respect, data augment can be an alternative way. However, different patterns of elements inside the event data sample enhance the difficulty for data augment. In this paper, we propose a novel DR event data augment method using conditional generative adversarial networks (CGAN). The proposed method innovatively transforms original event data into a new time series data named DR cost, which implicitly characterizes the relationship between response load and affecting factors. Then CGAN is utilized to learn the conditional probability distribution of DR cost data which is less complicated than DR event data. The synthetic DR cost data can not only be transformed into new DR event data but also be directly utilized for the price responsive demand in power system dispatch. The numerical results demonstrate that the generated DR event data capture almost the same probability distribution as realistic data. Experiment simulations further confirm that the real-time PRD pricing method based on synthetic DR cost data can guide consumers for the desired response load.