基于高斯混合密度网络的风电功率建模及预测  

Wind Power Modeling and Prediction Based on Gaussian Mixture Density Networks

作　　者：张耀聪 刘姝君[2] 杜小泽[1,2] 吴江波 ZHANG Yaocong;LIU Shujun;DU Xiaoze;WU Jiangbo(School of Energy Power and Mechanical Engineering,North China Electric Power University,Beijing 102206,China;School of Energy and Power Engineering Lanzhou University of Technology,Lanzhou 730050,China)

机构地区：[1]华北电力大学能源动力与机械工程学院,北京102206 [2]兰州理工大学能源与动力工程学院,兰州730050

出　　处：《工程热物理学报》2025年第4期1056-1066,共11页Journal of Engineering Thermophysics

基　　金：甘肃省教育厅“双一流”重点项目(No.GCJ2022-38);甘肃省高校产业支撑计划项目(No.2022CYZC-21);甘肃省重点研发计划项目(No.22YF7GA163)。

摘　　要：风电功率波动较大,准确的风电预测能为运维调度提供保障。常见的时间序列模型只适用于超短期预测,基于神经网络对机组功率转换特性进行建模,从而能根据风速预报进行日前预测。该数据驱动方法可改善理论功率计算方法因地貌、尾流等因素产生的偏差,神经网络对实际机组工作特性的描述更加准确。进一步建立高斯混合密度网络评估功率不确定性,模型通过训练自动学习分布参数,获得非先验的多元分布,能更好地反映数据内在结构。Wind power exhibits significant fluctuations,and accurate wind power prediction can provide a guarantee for operation and maintenance scheduling.Common time series models are only suitable for ultra-short-term forecasting.By modeling the power conversion characteristics of wind turbines using neural networks,it is possible to perform day-ahead predictions based on wind speed forecasts.This data-driven approach can improve the deviations caused by terrain,wake effects,and other factors in theoretical power calculation methods.Neural networks provide a more accurate description of the actual operating characteristics of wind turbines.Furthermore,a Gaussian Mixture Density Network(GMDN)is established to assess power uncertainty.The model automatically learns distribution parameters through training,resulting in a non-prior multivariate distribution that better reflects the inherent structure of the data.

关 键 词：风电预测 密度网络 功率建模 不确定性 

分 类 号：TK01[动力工程及工程热物理]