摘要
针对光伏输出功率预测问题,提出相似样本及PCA相结合的光伏输出功率预测模型。通过对光伏发电系统历史发电量数据和气象数据相关性分析,根据辐照度具有时间周期性和邻近相似性的特性选取参考样本,求取预测日与参考样本辐照度的欧氏距离并确定相似样本,采用PCA对相似样本提取主成分作为神经网络的输入,简化网络结构。仿真结果表明,相似样本算法可以有效地对不同天气类型的光伏输出功率进行预测,基于PCA的神经网络模型可进一步提高预测精度、泛化性能更好。
This article provides a photovohaic power output prediction model based on similar samples and principal component analysis. Correlation analysis is made according to the historical data of photovohaic power generation system＇ s electric energy production and meteorological data. Samples for reference are selected on the basis of solar radiation＇ s periodicity and adjacent similarity. Similar samples are obtained by calculating the Euclidean distance of the predicted date and the reference samples about solar radiation. PCA is used to abstract principal components of similar samples, which is the input of neural network. All these factors combined can eliminate the correlation of input variables and simplify the structure of network. The result of simulation presented that similar sample algorithm can effectively predict photovohaie power output under different kinds of weather. Neural network model based on principal component analysis can further improve the accuracy and have better generalizati
出处
《太阳能学报》
EI
CAS
CSCD
北大核心
2016年第9期2377-2385,共9页
Acta Energiae Solaris Sinica
基金
福建省自然科学基金(2015J01639)
福建省教育厅科技面上项目(JA15263)
福建省自然科学青年基金(2013J05081)
交通运输部科学技术研究计划(2015329815160)
