摘要
针对传统光伏功率预测特征提取不足导致预测精度不高的问题,提出一种双通道网络模型进行光伏功率预测。首先将光伏功率历史数据进行归一化处理,再将数据送入两个并行的卷积神经网络(Convolutional Neural Network,CNN)进行特征提取,经融合层融合送入长短期记忆网络(Long Short-Term Memory,LSTM)进行光伏功率预测。采用地中海气候光伏发电数据集进行测试,结果表明所提出的方法与单通道网络相比平均绝对误差(Mean-Absolute Error,MAE)减小了12.3%,均方根误差(Root-Mean-Square Error,RMSE)减小了3%,实现了更高的预测精度。
Aiming at the problem that the traditional PV power prediction feature extraction is insufficient and the prediction accuracy is not high, a two-channel network model is proposed for PV power prediction. Firstly, the PV power historical data is normalized, and then the data is sent to two parallel convolutional neural networks (CNN) for feature extraction. The fusion layer is used to fuse the features and then sent to the long-term and short-term memory network (LSTM) for PV power forecasting. The test was carried out using the mediterranean climate photovoltaic power generation dataset. The results show that the proposed method has a MAE reduction of 12. 3% and a RMSE reduction of 3% compared with the single-channel network, achieving higher prediction accuracy.
作者
简献忠
顾洪志
王如志
JIAN Xianzhong;GU Hongzhi;WANG Ruzhi(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;Shanghai Key Laboratory of Modern Optical System,Shanghai 200093,China;School of Mechanical Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;School of Materials Science and Engineering,Beijing University of Technology,Beijing 100020,China)
出处
《电力科学与工程》
2019年第5期7-11,共5页
Electric Power Science and Engineering
基金
国家自然科学基金项目(11774017)
关键词
光伏功率预测
深度学习
卷积神经网络
LSTM
photovoltaic power prediction
deep learning
convolutional neural network
LSTM
