基于相似日相关性聚类的LSTM短期光伏出力预测模型研究
Study on LSTM Short-term PV Output Prediction Model based on Similar Daily Correlation Clustering
杨堃 1安莉娜 1朱天生 1李大成 1庞锋 1项华伟1
作者信息
- 1. 中国电建集团贵阳勘测设计研究院有限公司,贵州省贵阳市 550081
- 折叠
摘要
为改善相似日聚类效果,提高光伏出力预测的准确性,提出了一种基于相似日相关性聚类的LSTM短期光伏出力预测模型.首先,计算各因素与光伏的相关性,筛选出与光伏出力相关程度较高的气象数据;然后,选取相关性系数分布均匀的光伏日出力过程作为初始聚类中心,以相关性为分类依据进行相似日聚类;最后,对不同聚类簇建立不同的长短期记忆神经网络(LSTM)训练模型.将相似日聚类与K-means聚类方法进行对比,结果显示相似日聚类的光伏出力预测均方根误差(RMSE)和平均绝对误差(MAE)分别降低了 19.2％和34.0％,决定系数R2提高了 4.8％,表明本文提出的基于相似日相关性聚类方法有效提高了光伏出力预测精度.
Abstract
In order to improve the effect of similar day clustering and improve the accuracy of PV prediction,a LSTM short-term PV output prediction model based on similar day correlation clustering was proposed.Firstly,the correlation between each factor and PV output was calculated to screen out the meteorological data with high correlation with PV output.Then,the daily PV output process with uniform correlation distribution is selected as the initial clustering center,and similar daily clustering is carried out based on correlation.Finally,different LSTM training models are built for different clusters.Comparing similar daily clustering with K-means clustering,The results show that the Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)of similar daily clustering are reduced by 19.2％ and 34.0％ respectively.The coefficient of determination R2 increased by 4.8％.In this paper,the LSTM short-term PV output prediction model based on similar daily correlation clustering proposed can effectively improve the accuracy of PV output prediction.
关键词
相似日/相关性/短期光伏出力预测/LSTMKey words
similar day/correlation/short-term prediction of photoltaic/LSTM引用本文复制引用
基金项目
中国电建集团贵阳勘测设计研究院有限公司科技项目(YJZD2022-01)
出版年
2024
NETL
NSTL
万方数据