摘要
针对分布式光伏出力预测精度提升和不确定性量化难题,提出了一种基于扩展长短期记忆网络(extended long short-term memory,xLSTM)和贝叶斯神经网络(bayesian neural network,BNN)的分布式光伏出力概率预测方法。首先,利用皮尔逊相关系数和最大信息系数对总辐照度、总云量、环境温度、风速等影响因素进行相关性分析,选取关键因素作为模型输入特征;然后,利用xLSTM进行时间序列特征提取,捕捉分布式光伏出力随时间变化的复杂波动模式;最后,将由xLSTM提取的有效特征输入BNN进行概率建模,从而量化分布式光伏出力的不确定性。基于中国某真实分布式光伏电站数据进行算例分析,结果表明所提概率预测方法与其他对比模型相比覆盖率平均提升了6.40%,区间宽度平均降低了16.82%。
To address the challenges of enhancing prediction accuracy and quantifying uncertainty in distributed photovoltaic(PV)power forecasting,this paper proposes a probabilistic forecasting method based on an Extended Long Short-Term Memory network(xLSTM)and Bayesian Neural Network(BNN).First,correlation analysis utilizing the Pearson correlation coefficient and Maximal Information Coefficient(MIC)is performed on influential factors including global horizontal irradiance(GHI),total cloud cover,ambient temperature,and wind speed to identify key features as model inputs.Then,the xLSTM is employed to extract temporal features,capturing the complex fluctuation patterns of distributed PV power output over time.Finally,the salient features extracted by the xLSTM are fed into the BNN for probabilistic modeling,thereby quantifying the uncertainty in distributed PV power generation.Case studies utilizing operational data from a distributed PV plant in China demonstrate that the proposed probabilistic forecasting method achieves an average improvement of 6.40%in prediction interval coverage probability(PICP)and an average reduction of 16.82%in the average prediction interval width(PINAW)compared with benchmark models.
作者
魏颖莉
栾士岩
贾亚飞
程鹏
张沛
WEI Yingli;LUAN Shiyan;JIA Yafei;CHENG Peng;ZHANG Pei(State Grid Xiong'an New Area Power Supply Company,Xiong'an New Area 071708,China;Collegeof Electrical Automation and Information Engineering,Tianjin University,Tianjin 300072,China)
出处
《河北电力技术》
2025年第5期1-8,共8页
Hebei Electric Power
基金
国网河北省电力有限公司科技项目(kj2024-058)。
关键词
分布式光伏
光伏发电功率预测
扩展长短期记忆网络
贝叶斯神经网络
概率预测
distributed photovoltaic
photovoltaic power generation forecast
extended long short-term memory
bayesian neural network
probabilistic prediction
