Generative Data-Driven Modeling of Aggregated Demand Curve in Day-Ahead Market

This paper proposes a Generative-Data-Attention-Unet based method to conduct short-term modeling of aggregate demand curves (ADCs) in the day-ahead market (DAM). We first formulate the ADC modeling from a machine vision perspective. To overcome the lack of historical data, we generate the bidding data based on the ADC functions with prior knowledge. These synthesized data and functions are visualized as images and paired as training samples. After training, the developed attention U-net is able to extract a figure of ADC from the processed input data. Last, the pixel mapping and polynomial fitting process are developed to derive a mathematical form of the output ADC figure. Numerical results show the superiority of the proposed method via benchmarks.