Graph Deep-Learning-Based Retail Dynamic Pricing for Demand Response

Abstract

Designing customized dynamic pricing is a promising way to incent consumers to adjust their daily en-ergy consumption behaviors. It helps manage flexible de-mand response resources on peak load. However, it is insuf-ficiently investigated in previous studies from the individual behavior perspective. To tackle the gap, this paper proposes a graph deep learning-based retail dynamic pricing mecha-nism. First, a graph attention network-based temporal price elasticity perceptron model is proposed. It explores a novel path to learn price elasticity by using graph deep learning, and can accurately assess consumers energy consumption behaviors under different prices. Then, to avoid unfair eval-uation of demand response, two indexes are proposed as auxiliary measures to assess energy consumption behavior learning models. At last, a customized dynamic pricing model based on the temporal price elasticity perceptron model is proposed. It can develop consumers time-varying demand response potential. This potential is first defined in this paper to measure what potentials of shifting/curtailing energy during a period a consumer has. By the pricing, the consumer could be incented to engage in demand response. The numerical studies validate the feasibility and superior-ity of the proposed methods, meanwhile price risks from the price change can be hedged effectively.