Heterogeneous graph-enhanced approach for demand response potential modeling: Mining load flexibility from user micro-behavioral patterns

Flexible resource scheduling on the electricity side is an important means to ensure the full consumption of renewable energy and the safe and stable operation of the power system. It is of great significance for the construction and development of new power systems. In the context of increasingly complex residential electricity consumption behavior, accurate load simulation and flexible load mining are important foundations for achieving demand response. Therefore, this article proposes a mining method for regional flexible loads that considers the micro electricity consumption patterns of power users. Firstly, through a double-layer clustering analysis based on multi-class fluctuation characteristics, accurately depict the electricity consumption patterns of power users. Secondly, considering the varying sensitivities of different households to environmental changes, a fine-grained electricity data simulation method is proposed that integrates heterogeneous graphs of user electricity consumption relationships to obtain refined energy consumption data. Finally, a flexible load mining method based on mixture density neural network is proposed, which mines regional flexible loads by combining the proportion of regional electricity consumption patterns. Taking the residential load in central China as an example, the proposed method achieves an overall simulation accuracy of 92.9 % and 90.5 % for residential load simulation on weekdays and holidays, respectively, which is higher than the simulation accuracy of traditional models; At the same time, this method successfully achieved load flexibility mining of regional residents on typical workdays and holidays, providing reliable technical support for flexible resource scheduling in the power system.