基于波动特性的新能源电力系统光伏净负荷预测方法

A Novel Photovoltaic Net Load Forecasting Method for New Energy Power Systems Based on Fluctuation Characteristics

中文关键词:波动特性  新能源电力系统  光伏净负荷预测  卷积神经网络结构  负荷特征  池化运算

keywords:fluctuation characteristics  new energy power system  photovoltaic net load forecasting  convolutional neural network structure  load characteristics  pooling operation

中文摘要:针对新能源电力系统光伏净负荷预测方法对特征提取不够精准，导致净负荷预测误差较大的问题，提出一种基于波动特性的新能源电力系统光伏净负荷预测方法。首先，对节点负荷数据进行剔除、缺失值补全以及归一化等预处理，量化影响因素变量与负荷间的关系；其次，将预处理后的数据进行量化聚类处理，根据不同的聚类结果，对光伏负荷特征进行精确提取，建立卷积神经网络结构；最后，对数据进行池化运算，代入负荷预测影响因素，对节点负荷序列进行分解、训练，输出池化运算后的预测数据，从而得到新能源电力系统光伏净负荷的预测结果。实验结果表明：所提方法可以对新能源电力系统光伏净负荷进行预测，且预测误差较小，其均方根误差约为3%，该方法具有较好的应用价值。

Abstract:To address the problem of insufficient accuracy in feature extraction of photovoltaic net load forecasting methods for new energy power systems， which results in large net load forecasting errors， a photovoltaic net load forecasting method for new energy power systems based on fluctuation characteristics is proposed. Firstly， preprocessing including outlier elimination， missing value imputation and normalization is performed on node load data to quantify the relationship between influencing factors and loads. Secondly， quantitative clustering is conducted on the preprocessed data， and photovoltaic load features are accurately extracted according to clustering results to construct a convolutional neural network structure. Finally， pooling operations are carried out on the data， load forecasting influencing factors are introduced， and the node load sequence is decomposed and trained. The predicted data after pooling operations are output to obtain the photovoltaic net load forecasting results of the new energy power system. Experimental results show that the proposed method can effectively forecast the photovoltaic net load of new energy power systems with small forecasting errors， and its root mean square error is about 3%， which verifies the favorable application value of the method.