基于模糊加权CNN的分布式光伏短期出力预测  

Distributed photovoltaic short-term output prediction based on fuzzy weighted CNN

作　　者：洪杨 王同乐 孔娟 王鹏 吴亚新 HONG Yang;WANG Tong-le;KONG Juan;WANG Peng;WU Ya-xin(Beijing China-Power Information Technology Company Limited,Beijing 100085,China)

机构地区：[1]北京中电普华信息技术有限公司,北京100085

出　　处：《信息技术》2025年第2期150-155,共6页Information Technology

摘　　要：为提升光伏短期出力预测准确性,提出基于模糊加权卷积神经网络的分布式光伏短期出力预测方法。筛选气象数据,按照数据特征值陡坡图,预处理数据,建立模糊加权卷积神经网络模型,将模糊卷积网络结构分为不同的层级,基于隶属度矩阵对数据进行最小归一化处理,按照模糊算法,求解最优值,根据预测因子,以粒子飞行算法为基础,按照水平斜面辐照度模型,量化天气数值并训练模型输出,从而实现光伏出力短期预测。实验结果表明,应用文中方法对分布式光伏短期出力进行预测,平均相对误差较小,为0.245。To improve the accuracy of short-term photovoltaic output prediction,a distributed photovoltaic short-term output prediction method based on fuzzy weighted convolutional neural network is proposed.The meteorological data is filtered,and according to the steep slope map of the data eigenvalues,the data is preprocessed.A fuzzy weighted convolutional neural network model is established,the fuzzy convolutional network structures are divided into different levels to perform minimum normalization on the data based on the membership matrix.Meanwhile,the optimal value is solved according to the fuzzy algorithm,and based on the prediction factor and particle flight algorithm,the weather values are quantified and the model output is trained according to the horizontal slope irradiance model,thus achieving short-term prediction of photovoltaic output.The experiment results show that the use of method proposed in this paper can predict the short-term output of distributed photovoltaics,and has a relatively small average relative error 0.245.

关 键 词：模糊加权卷积神经网络 分布式光伏 光伏出力 粒子飞行算法 隶属度矩阵 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]