基于气象融合与深度学习的分布式光伏出力区间预测
Prediction of Distributed Photovoltaic Output Interval Based on Meteorological Fusion and Deep Learning
葛亚明 1戴上 1梁文腾 1李言 1宋东阔 2陈金 2周霞 3单宇3
作者信息
- 1. 国网江苏省电力有限公司,南京 210008
- 2. 国电南瑞科技股份有限公司,南京 211000
- 3. 南京邮电大学,南京 210023
- 折叠
摘要
针对目前分布式光伏功率点预测方法难以全面的描绘分布式光伏功率出力不确定性等问题,提出了一种基于融合气象历史数据和WOA-BiLSTM的分布式光伏区间预测模型.首先,采用气象聚类方法,将历史数据集划分为晴天、多云等4个大类,得到包含各类转变天气过程的气象融合数据集.然后输入待预测日模糊天气类型,根据鲸鱼算法优化的双向长短期记忆网络模型得到光伏功率点预测值.最后通过核密度估计方法对点预测误差进行概率密度估计,叠加点预测值得到总体的预测区间结果.通过实际算例分析,与其他经典模型相比,所提方法具有更高的预测精度.
Abstract
As it is difficult to fully describe the uncertainty of distributed photovoltaic power output with the current distributed photovoltaic power point prediction method,a distributed photovoltaic interval prediction model based on meteorological historical data and WOA-BiLSTM is proposed in this paper.First,the meteorological clustering method is used to divide the historical data set into four categories:sunny,cloudy and so on,to obtain the meteorological fusion data set containing various types of transition weather processes.Second,the fuzzy weather type of the day to be predicted is input,and the PV power point prediction value is obtained according to the two-way long short-term memory network model optimized by the whale algorithm.Finally,the kernel density estimation method is used to estimate the probability density of the point prediction error,and the overall prediction interval result is obtained by superimposing the point prediction value.Through the analysis of practical examples,it is found that compared with other classical models,the proposed method has higher prediction accuracy.
关键词
气象监测装置/分布式光伏/点预测/深度学习/融合气象信息/鲸鱼算法/核密度估计/区间预测Key words
meteorological monitoring device/distributed photovoltaic/point prediction/deep learning/fusion of meteorological information/Whale algorithm/kernel density estimation/interval prediction引用本文复制引用
基金项目
国家自然科学基金重点项目(61933005)
出版年
2024
NETL
NSTL
万方数据