王帅 杜欣慧 姚宏民
摘 要: 密度峰值聚类算法具有收敛速度快、鲁棒性强、无需人为确定最佳聚类数等特点,具备较好的应用前景。为提高光伏功率预测的精度,提出一种将密度峰值聚类算法应用于短期光伏功率预测的方法,并进行了必要优化。该方法首先通过类间距离优化增强气象数据的可分性;然后利用密度峰值聚类对其进行无标签归类,通过灰色关联度匹配出与待预测日相关度最高的类别;最后将其作为Elman神经网络的训练样本,得到预测结果。Matlab仿真结果表明,该方法能够明显提高气象数据的聚类效果,并有效提高光伏功率的短期预测精度。
关键词: 密度峰值聚类; 光伏发电; 灰色关联度; 相似日匹配; Elman神经网络; 短期功率预测
中图分类号: TN830.4?34; TM715 文献标识码: A 文章编号: 1004?373X(2018)20?0141?05
Abstract: The density peak clustering algorithm has the characteristics of fast convergence speed, strong robustness, and no need to manually determine the optimal clustering numbers, which has a good application prospect. Therefore, a method of applying the density peak clustering algorithm to short?term photovoltaic power prediction is proposed and necessarily optimized to improve the accuracy of photovoltaic power prediction. In the method, the separability of meteorological data is enhanced by means of inter?class distance optimization. The density peak clustering is adopted to classify the meteorological data without using labels. The category that has the highest degree of correlation with to?be predicted days is matched by using the grey correlation degree, and finally taken as the training sample of the Elman neural network to obtain the prediction results. The Matlab simulation results show that the method can significantly improve the clustering effect of meteorological data and short?term prediction accuracy of photovoltaic power.
Keywords: density peak clustering; photovoltaic power generation; grey correlation degree; similar day matching; Elman neural network; short?term power prediction
光伏发电系统因其电能可控性强,调度管理便捷、调峰性能良好、响应迅速等优点而得到了广泛应用。但由于其输出功率受风速、气温、阳光辐射度、日降雨量、气压等气象因素影响存在较大随机波动性[1?3],致使有功功率不稳定而对电网造成干扰。因此,精准的光伏发电功率预测有助于降低对电网的影响,也便于电力工作人员的调度管理工作。
目前,采用较多的预测方法是神经网络预测法[4?5]。即使用历史发电量与气象数据组成的训练样本训练神经网络的自学习能力,将待预测日的气象特征(通过数值天气预报获得)输入网络后得到预测结果。其中,提高训练样本与待预测日的相关性极其重要。……