模糊聚类-Elman神经网络短期光伏发电预测模型

Fuzzy clustering-Elman neural network model for short-term photovoltaic power generation prediction

光伏发电功率预测是电网安全稳定运行的基础,从数据挖掘的角度提升光伏发电功率预测精度,文中提出基于孤立森林、模糊C均值和Elman的短期光伏发电功率预测模型。根据预测日选择相似日数据并按天气分类作为训练样本;采用孤立森林清洗训练样本中的异常部分;应用模糊C均值对相似日以及待预测日的气象数据进行聚类分析。结合Elman神经网络算法,形成含孤立森林数据清洗的模糊聚类-Elman神经网络的预测模型,对光伏出力进行精确预测。根据某地市现场实测数据进行实验仿真,预测结果分别与传统Elman和BP模型的预测结果进行对比,所述模型可以获得更高的预测精度。

Photovoltaic power generation forecasting is the basis of safe and stable operation in power grids.The accuracy of PV power generation prediction is improved from the perspective of data mining.This paper proposes a short-term photovoltaic power generation prediction model based on Isolation Forest,fuzzy C means and Elman.Firstly,similar daily data are selected according to the forecast date and classified according to the weather as training samples.Secondly,the abnormal parts in the Isolation Forest cleaning training samples are adopted.Thirdly,fuzzy C means clustering method is utilized to cluster the meteorological data of similar days and the forecasting days.Finally,the Elman neural network algorithm is applied to form a prediction model with Isolation Forest data cleaning method based on fuzzy clustering-Elman neural network to the accurate forecasting of the PV output.The experimental simulation is carried out according to the actual measured data of a certain city in Anhui province.The prediction results are respectively compared with the traditional Elman and BP models.It is demonstrated that higher prediction accuracy can be obtained.