| 引用本文: | 廖文龙,任翔,杨哲,杨文清,魏超.基于隐式最大似然估计的风电出力场景生成[J].电力自动化设备,2022,42(11): |
| LIAO Wenlong,REN Xiang,YANG Zhe,YANG Wenqing,WEI Chao.Scenario generation of wind power output based on implicit maximum likelihood estimation[J].Electric Power Automation Equipment,2022,42(11): |
|
| 摘要: |
| 随着风电渗透率的日益提高，如何有效地描述风电出力的不确定性成为了配电网运行和规划所面临的巨大挑战，为此，提出一种基于隐式最大似然估计的风电出力场景生成方法。针对风电出力曲线的数据特征，设计适用于风电出力场景生成的损失函数和网络结构。通过无监督训练使得场景生成器能够学习到高斯噪声与风电出力场景之间的映射关系。仅需调节模型中相关的参数，采用所提方法就能够生成不同时间尺度的风电出力场景。仿真结果表明，所提方法的预测区间平均宽度和预测区间覆盖率均优于现有的生成对抗网络，且所提方法对于不同的风电场具有一定的普适性。 |
| 关键词: 风电 场景生成 生成模型 深度学习 隐式最大似然估计 |
| DOI：10.16081/j.epae.202205006 |
| 分类号:TM614 |
| 基金项目: |
|
| Scenario generation of wind power output based on implicit maximum likelihood estimation |
|
LIAO Wenlong1, REN Xiang2, YANG Zhe1, YANG Wenqing3, WEI Chao4
|
|
1.Department of Energy, Aalborg University, Aalborg 9220, Denmark;2.Electric Power Research Institute of State Grid Jibei Electric Power Co.,Ltd.,Beijing 100045, China;3.Department of Statistics, Columbia University, New York NY10025, USA;4.Guodian Northeast New Energy Development Co.,Ltd.,Shenyang 110000, China
|
| Abstract: |
| With the increasing penetration of wind power, how to effectively describe the uncertainty of wind power output has become a huge challenge for the operation and planning of distribution network, for which, a scenario generation method of wind power output is proposed based on implicit maximum likelihood estimation. According to the data characteristics of wind power output curves, the loss function and network structure suitable for scenario generation of wind power output are designed. Through unsupervised training, the scenario generator can learn the mapping relationship between Gaussian noise and wind power output scenarios. The wind power output scenarios with different time scales can be generated with the proposed method by only adjusting the relevant parameters in the model. The simulative results show that both the forecasting interval average width and forecasting interval coverage percentage of the proposed method are better than those of the existing generative adversarial network, and the proposed method has certain universality for different wind farms. |
| Key words: wind power scenario generation generative model deep learning implicit maximum likelihood estimation |
