基于PCA和ML-ELM-AE的短期光伏功率预测  

Short-term Photovoltaic Power Prediction Based on PCA and ML-ELM-AE

作　　者：靳果[1,2] 朱清智 闫奇[3] JIN Guo;ZHU Qing-zhi;YAN Qi(Electromechanical Automation College,Henan Polytechnic Institute,Nanyang 473000,China;Institute of Photonics&Photon-technology,Northwest University,Xi'an 710000,China;State Grid Nanyang Power Supply Company,Nanyang 473000,China)

机构地区：[1]河南工业职业技术学院机电自动化学院,河南南阳473000 [2]西北大学光子学与光子技术研究所,陕西西安710000 [3]国网南阳供电公司,河南南阳473000

出　　处：《控制工程》2021年第9期1787-1796,共10页Control Engineering of China

基　　金：河南省2020年科技发展计划项目(202102210134);河南省2021年科技发展计划项目(212102210527);河南工业职业技术学院青年骨干教师培养计划项目(202007)。

摘　　要：光伏发电系统输出功率受气象因素季节性和随机性的影响,表现出明显的间歇性和波动性。为了提高各种情况下的输出功率预测精度,提出了一种基于主成分分析和自编码结构多层极限学习机组合算法的预测模型。首先通过相关性分析明确影响输出功率的主要气象因素,然后根据相似日原则选取待测日的训练样本和测试样本,最后利用光伏电站历史气象数据和输出功率数据对预测模型进行训练和测试,验证预测方法的预测效果。实验结果表明,与反向传播神经网络算法和支持向量机算法相比,所提出的预测方法在不同季节和不同天气情况下均具有较高的预测精度,表现出良好的功率预测稳定性和泛化能力。Influenced by the seasonality and randomness of meteorological factors, The output power of photovoltaic power generation system shows obvious intermittence and fluctuation. In order to improve the prediction accuracy of output power in various conditions, an output power prediction model based on principal component analysis(PCA) and multi-layer extreme learning machine auto encoder(ML-ELM-AE) combined algorithm is proposed. Firstly, the main meteorological factors affecting the output power are identified through correlation analysis. Secondly, the training samples and testing samples of the tested day are selected according to the similarity day principle. Finally, the prediction model is trained and tested by using the historical meteorological datasets and output power datasets of a photovoltaic power plant to verify the performance of the prediction method. The experimental results show that, compared with back propagation neural network algorithm and support vector machine algorithm, the proposed method has higher prediction accuracy in the case of different seasons and weather, and excellent power prediction stability and generalization ability.

关 键 词：相似日 主成分分析 极限学习机 季节 天气 

分 类 号：TM615[电气工程—电力系统及自动化]