Forecasting electricity demand based on weather effects using diffusion model and causal attention

Short-term electricity load forecasting is a critical component of power system dispatch operations. As a core variable influencing load fluctuations, the precise quantification of meteorological factors impact has long been a research challenge. This paper proposes an innovative forecasting method integrating a diffusion model with a causal attention mechanism. This approach utilises the diffusion model to capture the randomness and uncertainty inherent in meteorological factors, while explicitly modelling the causal relationship between weather variables and electricity load through the causal attention mechanism. Experiments on public datasets demonstrate that the proposed method reduces prediction errors by 12% compared to traditional long short-term memory models, achieving over 90% prediction accuracy during extreme weather events. This provides a new pathway for refining the quantification of meteorological impacts and offers significant reference value for power system dispatch decision-making.