A Novel Algorithm for Load Prediction under Atypical Climate based on Multi-source Data Fusion

Accurate prediction of household load plays a crucial role in optimizing power distribution and enhancing energy efficiency. Household loads are subject to a multitude of uncertainties, especially for climatic conditions. To understand the differential impacts of climatic factors on household load, this research undertakes a comprehensive in-depth investigation of the effects of typical and atypical climate conditions on household load. In this paper, a hybrid predictive model integrating Long Short-Term Memory (LSTM) and Self-Attention mechanisms is put forward for the hourly prediction of residential load under different climate conditions. We integrate the typical climate and the atypical climate data into the household load and analyze their effects on the household load respectively. The experimental results substantiate that the novel hybrid architecture outperforms the standalone LSTM model in household load prediction when combined with climate data. Therefore, incorporating typical and atypical climate data into household load predation can enhance forecasting accuracy.