基于卷积神经网络和卫星云图空间权重的光伏功率估计研究

Study of photovoltaic power based on convolutional neural network and spatial weights of satellite cloud maps

摘要:

本文研究了一种基于卷积神经网络和卫星云图空间权重的光伏功率回归重构方法。为了更全面精确地反映太阳直射和太阳散射对光伏功率的影响,本文提出了一种对卫星云图进行赋权重处理方法,并据此构建了一种新型的光伏功率回归重构方法。该模型首先对卫星云图进行处理以增强云和背景的区别,随后利用历史功率与云图数据,计算了不同方位角下光伏功率与不同空间位置卫星云图的相关系数。最后,利用卷积神经网络模型建立计及空间权重的卫星云图与光伏功率之间的映射关系,实现光伏功率回归重构。实验结果表明,所提模型通过赋权重处理充分考虑了太阳直射和太阳散射的影响,并展现出了良好的精度。本研究为基于卫星云图的光伏功率回归重构提供了有益的参考。

Abstract:

In this paper, a regression reconstruction method of photovoltaic power based on convolutional neural network and spatial weights of satellite cloud map is investigated. In order to more comprehensively and accurately reflect the effects of direct sunlight and solar scattering on PV power, this paper proposes a method of assigning weights to satellite cloud maps, and constructs a novel PV power regression reconstruction method accordingly. The model firstly processes the satellite cloud map to enhance the difference between cloud and background, and then the correlation coefficients between PV power at different azimuths and satellite cloud maps at different spatial locations are calculated using historical power and cloud map data. Finally, a convolutional neural network model is used to establish the mapping relationship between satellite cloud maps and PV power taking into account the spatial weights to realize the regression reconstruction of PV power. The experimental results show that the proposed model fully considers the effects of direct sunlight and solar scattering through the weighting process, and demonstrates good accuracy. This study provides a useful reference for the regression reconstruction of PV power based on satellite cloud maps.