基于宽度学习系统的光伏发电功率超短期预测

近年来,深度学习被应用于光伏发电预测中,体现出预测精度较高的优点,但也存在训练耗时等问题。对此,提出了一种基于自组织映射与宽度学习系统的光伏发电功率超短期预测模型。首先,采用自组织映射对各时刻的光伏数据进行精细化聚类,提取不同时段与气象条件下的出力波动特征;然后,构建基于宽度学习系统的光伏发电功率多步长预测模型,在网络宽度上扩展节点数目,并通过求解矩阵伪逆训练神经网络,在保证较强高维数据拟合能力的同时,具有较高的计算效率;最后,采用实际光伏发电数据进行算例分析,通过与常用的光伏发电超短期预测方法进行比较,验证所提出的方法在预测精度与训练效率上的优越性。

Ultra-short-term Forecasting of Photovoltaic Power Generation Based on Broad Learning System

In recent years, deep learning has been applied to photovoltaic (PV) power forecasting and shows the advantage of high accuracy. However, there also exist problems such as time-consuming training. In order to handle these problems, an ultra-short-term forecasting model of PV power generation based on the self-organizing map (SOM) and broad learning system (BLS) is proposed. Firstly, the SOM is applied to cluster the PV data at each moment to extract the fluctuation characteristics during different periods and under different meteorological conditions. Secondly, a multi-step forecasting model of PV power generation based on BLS is constructed, which increases the number of neurons in width and is trained by calculating the pseudoinverse matrix, thus guaranteeing strong capability for fitting high-dimensional data while keeping high computation efficiency. Finally, the actual PV power generation data are used to carry out the case study. By comparing with commonly used ultra-short-term forecasting methods, the superiority of the proposed method in terms of forecasting accuracy and training efficiency is verified. This work is supported by National Natural Science Foundation of China (No. 52077136).