摘要
针对光伏功率数据稳定性低、波动性大以及通过单一模型难以全面捕捉信号非线性特征的问题,提出了一种基于二次分解和JSO-TCN模型的光伏预测模型。该模型首先通过自适应噪声完备集合经验模态分解(CEEMDAN)对实际光伏功率数据进行分解;然后分别计算各分量的样本熵,并通过K-means++聚类为高频、中频和低频3个分量,再利用变分模态分解(VMD)对熵值最高的模态分量进行二次分解;最终将处理后的数据输入到时序卷积网络(TCN)中并采用水母优化算法(JSO)对TCN进行参数优选。以西南地区某光伏电站为例,相比于其他模型,本模型在3类指标上均具有优势,决定系数(R 2)为98.29%、平均绝对误差(MAE)为0.481 MW、均方根误差(RMSE)为0.674 MW。由此可知,基于二次分解和JSO-TCN模型预测精度高、误差小,能够为该地区电网调度提供参考。
This paper proposes a photovoltaic prediction model based on secondary decomposition and JSO-TCN model to address the issues of low stability,high volatility,and difficult to capture signal nonlinearity comprehensively with a single model for photovoltaic power data.The model first decomposes actual photovoltaic power data using complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN),then calculates the sample entropy of each component,and clusters them into three components of high frequency,medium frequency and low frequency using K-means++clustering.Next,it performs secondary decomposition on the mode component with the highest entropy using variational mode decomposition(VMD).The processed data is finally inputted into a temporal convolutional network(TCN),and the TCN is optimized using the jellyfish swarm optimization algorithm(JSO).Using a photovoltaic power station in the Southwest region as an example,compared to other models,this model has advantages in three categories of metrics,with a coefficient of determination(R 2)of 98.29%,mean absolute error(MAE)of 0.481 MW,and root mean square error(RMSE)of 0.674 MW.Therefore,it can be concluded that the secondary decomposition-based JSO-TCN model has high prediction accuracy and low error,providing valuable insights for grid dispatching in the region.
作者
钟璐
杨华
李世林
亢丽君
马光文
朱燕梅
黄炜斌
ZHONG Lu;YANG Hua;LI Shilin;KANG Lijun;MA Guangwen;ZHU Yanmei;HUANG Weibin(State Grid Southwest Division,Chengdu 610041,Sichuan,China;College of Water Resources and Hydropower,Sichuan University,Chengdu 610065,Sichuan,China)
出处
《水力发电》
CAS
2024年第11期74-80,105,共8页
Water Power
基金
国家电网有限公司西南分部科技项目(SGSW0000D-KJS2310037)。
关键词
光伏功率
预测
自适应噪声完备集合经验模态分解
变分模态分解
样本熵
K-means++聚类
水母优化算法
时序卷积网络
photovoltaic power
prediction
completely empirical mode decomposition with adaptive noise
variational mode decomposition
sample entropy
K-means++clustering
jellyfish search optimizer
temporal convolutional networks
