摘要
风电功率预测结果的不确定性包括数据不确定性和模型不确定性。首先,分析了两种不确定性的来源,给出了风电功率概率预测的目标和形式;其次,将风电功率条件概率分布的参数作为神经网络输出,并利用负对数似然损失作为损失函数,实现对数据不确定性的建模;然后,将神经网络的权重由确定的变量转变为随机变量,并采用概率分布表示,实现了对模型不确定性的建模;最后,提出了考虑数据不确定性和模型不确定性的风电功率概率预测方法。基于实际风电场数据分析了不同概率分布下预测的性能,结果表明,所提方法支持概率分布、区间、出力场景3种形式概率预测;并且考虑模型不确定性后,提升了概率分布预测的性能;也验证了在异常天气条件下,所提方法能够表征预测结果的不确定性。
The uncertainty of wind power prediction results includes data uncertainty and model uncertainty.Firstly,the sources of the two uncertainties are analyzed,and the objectives and forms of wind power probability prediction are given;Secondly,the parameter of the probability distribution of wind power is used as the neural network output,and the negative log likelihood loss as the loss function to model the data uncertainties;Then,the weight parameters of neural network are taken as random variables and expressed by probability distribution to realize the modeling of the model uncertainties;Finally,based on the model,a wind power probability distribution prediction method considering the data uncertainties is proposed.The data of actual wind farms are used for verification,and the prediction performances under different probability distribution assumptions are analyzed.The results show that the method supports three forms of probability prediction:the probability distribution,the interval and the output scenarios.Considering the model uncertainties,the performance of probability distribution prediction is improved.It is also verified that the method can characterize the uncertainty of prediction results under abnormal weather conditions.
作者
余建明
庞传军
YU Jianming;PANG Chuanjun(NARI Group Corporation Co.,Ltd.(State Grid Electric Power Research Institute Co.,Ltd.),Nanjing 211106,Jiangsu Province,China;Beijing KeDong Electric Power Control System Co.,Ltd.,Haidian District,Beijing 100192,China)
出处
《电网技术》
EI
CSCD
北大核心
2022年第5期1926-1933,共8页
Power System Technology
基金
国家电网有限公司科技项目(5700-202028362A-0-0-00)。
