基于时序混合密度网络的超短期风电功率概率预测  

Ultra-short-term Probabilistic Forecasting of Wind Power Based on Temporal Mixture Density Network

作　　者：董骁翀 孙英云[1] 蒲天骄 王新迎 李烨 DONG Xiaochong;SUN Yingyun;PU Tianjiao;WANG Xinying;LI Ye(School of Electrical and Electronics Engineering,North China Electric Power University,Beijing 102206,China;China Electric Power Research Institute,Beijing 100192,China)

机构地区：[1]华北电力大学电气与电子工程学院,北京市102206 [2]中国电力科学研究院有限公司,北京市100192

出　　处：《电力系统自动化》2022年第14期93-100,共8页Automation of Electric Power Systems

基　　金：国家重点研发计划资助项目(2020YFB0905900);国家自然科学基金资助项目(51777065)。

摘　　要：风电概率预测能够为新型电力系统安全运行提供关键的边界条件。提高预测精度是风电概率预测研究的关键问题,并且提高隐式模型的可解释性有益于人工智能模型的推广应用。因此,文中提出了时序混合密度网络,提取风电时序数据的局部矩信息作为输入通道,采用时序卷积网络提取多时间尺度的概率特征,并使用混合Beta分布构建概率预测信息。算例结果表明,局部矩通道能有效提高模型训练的收敛性,并且由时序混合密度网络提取的混合分布参数具有一定的可解释性,其预测结果相比现有模型具有更高的精度。Probabilistic forecasting of wind power can provide critical boundary conditions for the safe operation of new power systems. Improving forecasting accuracy is the key problem of the research for probabilistic forecasting of wind power, and improving the interpretability of implicit models is beneficial to the promotion and application of artificial intelligence models.Therefore, a temporal mixture density network is proposed, which extracts the local moment information of time series data of wind power as input channels. The temporal convolutional network is used to extract the multi-time-scale probabilistic features,and the mixed Beta distribution is used to construct the probabilistic forecasting information. The results of the case study show that the local moment channel effectively improves the convergence of the model training, and the mixed distribution parameters extracted by the temporal mixture density network have a certain interpretability. Compared with the existing models, the forecasting results of the temporal mixture density network have better accuracy.

关 键 词：风电功率 概率预测 混合密度网络 时序卷积网络 最大似然估计 可解释性 

分 类 号：TM614[电气工程—电力系统及自动化]