基于天气分型的短期光伏功率组合预测方法

由于光伏功率波动特征与天气类型紧密相关,且光伏功率短期预测存在功率波动过程预测精度低、气象因素与功率波动过程相关性弱的问题,文中提出了一种基于天气分型的短期光伏功率组合预测方法。首先,基于气象因素与光伏功率波动特征的关联性,将天气过程划分为5种类型,并基于变分模态分解算法将光伏功率分解为类晴空过程和波动过程。然后,利用Granger因果关系算法筛选出与各天气类型下光伏功率波动过程密切相关的关键气象因子。最后,建立基于天气分型的短期光伏功率组合预测模型。模型充分考虑了深度学习算法的特异性,对光伏功率类晴空过程与各天气类型下的光伏功率波动过程进行分类预测。仿真结果表明,文中所提出的短期光伏功率预测方法能够显著提升短期光伏功率预测的精度。

Combination Forecasting Method of Short-term Photovoltaic Power Based on Weather Classification

The fluctuation characteristics of the photovoltaic (PV) power are closely related to the weather types, and the short-term PV power forecasting has problems of low forecasting accuracy in the power fluctuation process and the weak correlation between meteorological factors and the power fluctuation process. This paper proposes a combination forecasting method of short-term PV power based on weather classification. Firstly, the weather process is divided into five types based on the meteorological factors and fluctuation characteristics of PV power. Based on the variational mode decomposition algorithm, the PV power is decomposed into the clear-sky-like process and the fluctuation process. Secondly, the Granger causality algorithm is used to select the key meteorological factors, which are closely related to the fluctuation process of PV power with various weather types. Finally, a combined forecasting model of short-term PV power based on weather classification is established. The model fully considers the specificity of the deep learning algorithm, separately forecasts the clear-sky-like process and the fluctuation process of PV power with various weather types. The simulation results show that the proposed short-term PV power forecasting method can significantly improve the accuracy of the short-term PV power forecasting.