A Power-Forecasting Method for Geographically Distributed PV Power Systems using Their Previous Datasets
A Power-Forecasting Method for Geographically Distributed PV Power Systems using Their Previous Datasets
复制标题
DOI：
10.3390/en12244815
复制
发表时间：
2019-12
期刊：
影响因子：
3.2
通讯作者：
Y. Miyazaki;Yusuke Kameda;J. Kondoh
中科院分区：
文献类型：
--
作者：
Y. Miyazaki;Yusuke Kameda;J. Kondoh
全球正在安装的光伏(PV)电力系统的数量一直在增加。这使得维持供需平衡成为电力系统运营者非常关心的问题。预测光伏发电量是一种前景看好的对策，已经引起了人们的极大兴趣。实现这一点的传统方法通常使用学习方法，如神经网络和支持向量回归。相反，本文提出了一种适用于地理分布的光伏系统的短期功率预测方法，该方法仅使用其先前的输出功率数据。在该方法中，首先求出每个光伏系统在特定日期和时间的指定时段内每个观测实例的发电量与最大功率输出值的比值。然后，根据前一分布的时间变化(运动)预测该比率的未来地理分布。最后，将预测的功率比重新转换为功率输出，进行短期功率预测。对日本关东地区光伏总输出功率的预测结果表明，该方法的平均绝对百分比误差为4.23%，均方根误差为0.69 kW，验证了该方法的有效性。
The number of photovoltaic (PV) power systems being installed worldwide has been increasing. This has resulted in maintenance of an adequate balance between demand and supply becoming a great concern for power system operators. Forecasting PV power outputs is a promising countermeasure that has been garnering significant interest. Conventional methods for achieving this often use learning methods, such as neural networks and support vector regression. In contrast, this paper proposes a short-term power-forecasting method for geographically distributed PV systems that uses only their previous output power data. In the proposed method, first, the ratio of the power generation output to the maximum power output value for each observation instance in a designated period for each PV system at a certain date and time is obtained. Then, the future geographical distribution of the ratio is predicted from the temporal change (motion) of the preceding distribution. Finally, the predicted ratio is reconverted into the power output to perform short-term power forecasting. The results of total PV output power prediction in the Kanto area of Japan indicate that the proposed method has an average mean absolute percentage error of 4.23% and root mean square error of 0.69 kW, which verifies its efficacy.