基 于 条 件 增 强 扩 散 模 型 的 短 期 负 荷 概 率 预 测

Load probabilistic forecasting can provide guidance for power grid planning, and the conditional generation model can effectively improve the forecasting performance by mining historical similar-day information. However, previous studies did not pay attention to the curve shape information and the noise analysis function of unconditional models, which increased the uncertainty of the generation curve. Therefore, a short-term load probabilistic forecasting method based on conditional enhanced diffusion model is proposed. Firstly, an improved iTransformer daily load forecasting model is constructed to forecast the adjacent daily load data. Secondly, a diffusion model combining multi-head self-attention mechanism and U-net is constructed using a loss function that combines unconditional noise estimation and conditional noise estimation. Then, the daily load forecasting results and characteristics such as temperature are used as conditional inputs. Through the reverse diffusion process of conditional enhanced guidance, multiple sets of random noise are denoised to generate multiple load curves for probability density analysis. Finally, based on a publicly available dataset from a region in China and comparative tests with various models, the case study analysis demonstrates that the proposed method has higher forecasting accuracy.