Residential Electricity Load Scenario Prediction Based on Transferable Flow Generation Model

Day-ahead residential load forecasting is important for power system demand response. Considering the fluctuation of the residential electricity load and the small accumulation of electricity consumption data in some households, the prediction accuracy of the residential electricity consumption load is significantly challenging. In this study, a scenario prediction scheme for residential electricity consumption load using a transferable flow-based generation model was proposed. First, to make full use of the source domain data, different source domain families were selected to form multi-source domain families according to the association index of the source and target domains by introducing grey correlation analysis. Thereafter, the method of model transfer was adopted, and the pretraining model was established using multi-source household electrical load data. The network parameters of part of the step of flow structure were frozen in the pretraining model, the structural parameters of the unfrozen step of flow structure were fine-tuned and trained by household electrical load data in the target domain, and the day-ahead electricity load prediction model under a small sample was constructed. The experimental results show that the algorithm combined with model transfer performs well in solving the residential load-forecasting effect for small samples.

Automated summary

In this study, a scenario prediction scheme for residential electricity consumption load using a transferable flow-based generation model was proposed. Experimental results show that the algorithm performs well in solving the residential load-forecasting effect for small samples.