Multi-Information Fusion-Based Price Elasticity Evaluation of Flexible Load Demand via Machine Vision

Residential demand response (DR) assumes an increasingly crucial role in maintaining the stable operation of new power system. A pivotal element of residential DR is to figure out the hourly price elasticity of demand (PED) among consumers, which proves to be a challenging task due to the volatile electricity consumption patterns (ECP). To overcome this issue, a three-step methodology is proposed. Firstly, a parallel-computing and merging (PC&M) algorithm is proposed to categorize the massive residential ECPs based on historical data. Then, the category-specific ECPs are converted into 24 visual images to characterize the relationship between hourly load demand and corresponding electricity prices. Secondly, a demand curve (DC) function set is constructed based on prior knowledge. Moreover, the ground-truth data points are generated around the DC functions, which along with DC functions are paired as training samples for the Attention U-net based model via machine vision. Thirdly, by inputting hourly images of each representative ECP into the trained model, the polynomial sets for evaluating the 24-hour PED are obtained by the processes of pixel mapping and polynomial fitting. At last, a personalized encapsulation model is developed to assess ECP of each specific user and to exploiting its RDR potential. The numerical results indicate that the proposed method and model are effective and superior compared to the benchmark test.