基于CNN-GRU与特征增强的超短期光伏功率预测  

Ultra-Short-Term Power Prediction of PV Based on CNN-GRU with Feature Augmentation

作　　者：李宇豪 杨建卫 李佳瑞 刘永生 LI Yu-hao;YANG Jian-wei;LI Jia-rui;LIU Yong-sheng(Solar Energy Research Institute,Shanghai Electric Power University,Shanghai 201306,China;China Power Hua Chuang(Suzhou)Electricity Technology Research Co.,Ltd.,Suzhou Jiangsu 215123,China)

机构地区：[1]上海电力大学太阳能研究所,上海201306 [2]中电华创(苏州)电力技术研究有限公司,江苏苏州215123

出　　处：《计算机仿真》2024年第10期83-88,共6页Computer Simulation

基　　金：国家自然科学基金(51971128,52171185);上海市优秀学术/技术带头人计划(20XD1401800)。

摘　　要：超短期光伏功率预测对电力系统的实时调度有着重要意义。针对以往深度学习预测光伏输出功率重模型轻特征的特点,提出了一种基于CNN-GRU与特征增强的超短期光伏功率预测方法。首先将历史数据按照季节划分,以平抑季节性变化对光伏输出功率的影响。然后将可测数据基于其物理性质进行特征增强,使其能够被神经网络模型更充分的挖掘。最后采用CNN-GRU模型充分挖掘数据的时间与空间特征,进一步提升预测准确率。应用中国江苏某装机容量为75 MW光伏电站实际生产数据进行仿真验证,结果表明,上述方法在不同季节、天气情况下的预测精度均有较为明显的提升。Ultra-short-term PV power prediction is significant for the real-time dispatching of power systems.A ultra short term photovoltaic power prediction method based on CNN-GRU and feature enhancement is proposed to address the characteristics of previous deep learning models that emphasize models and underestimate features in predicting photovoltaic output power.Firstly,historical data are divided by seasons to mitigate the impact of seasonal variations on PV output power.Then the measurable data are feature-enhanced based on their physical properties so that they can be more fully exploited by the neural network model.Finally,the CNN-GRU model is used to fully exploit the temporal and spatial characteristics of the data to further improve the prediction accuracy.The simulation is validated by applying the actual production data of a 75 MW installed capacity PV plant in Jiangsu,China,and the results show that the prediction accuracy of the method is improved more significantly under different seasons and weather conditions.

关 键 词：超短期光伏功率预测 特征增强 倾斜辐照度 光伏电池温度 卷积神经网络 门控循环单元网络 

分 类 号：TM743[电气工程—电力系统及自动化]