基于条件去噪扩散隐式模型的光伏日前场景生成方法OA
Scenario generation method for day-ahead photovoltaic power based on a conditional denoising diffusion implicit model
为准确刻画光伏能源出力的不确定性,同时针对去噪扩散概率模型存在的缺陷,提出一种基于条件去噪扩散隐式模型的光伏日前场景生成方法.通过改进模型逆向去噪过程,实现确定性采样,将日前预测功率作为条件信息,引导逆向去噪过程生成符合条件分布的光伏出力场景;同时采用滑动窗口交叉验证策略,提升模型对时序特征的学习和泛化能力.仿真结果表明,所提方法能有效提升生成效率,且在高采样步数时生成场景能较好地反映光伏出力的时序相关性与概率分布特性,验证了其在新能源不确定性表征方面的有效性与实用性.
To accurately characterize the uncertainty of photovoltaic(PV)power output and address the limitations of denoising diffusion probabilistic models(DDPM),a day-ahead PV scenario generation method based on a condi-tional denoising diffusion implicit model(CDDIM)is proposed.The method modifies the reverse denoising process to enable deterministic sampling,where day-ahead forecasted PV power is incorporated as conditional information to guide the reverse diffusion process and generate PV power scenarios consistent with the conditional distribution.In addition,a sliding-window cross-validation strategy is adopted to enhance the model's capability in learning tem-poral features and improve its generalization performance.Simulation results demonstrate that the proposed method effectively improves generation efficiency.Under a higher number of sampling steps,the generated scenarios can well capture the temporal correlations and probabilistic distribution characteristics of PV power output,verifying the effectiveness and practical applicability of the method in representing renewable energy uncertainty.
毛夏平;罗萍萍;姚海佳
上海电力大学 电气工程学部,上海 200090上海电力大学 电气工程学部,上海 200090上海电力大学 电气工程学部,上海 200090
光伏能源不确定性场景生成条件去噪扩散隐式模型
photovoltaic poweruncertaintyscenario generationconditional denoising diffusion implicit model
《浙江电力》 2026 (9)
56-66,11
国家自然科学基金(51177107)
评论