基于自适应k均值和SVR的光伏出力预测

Photovoltaic Output Prediction Based on Adaptive k-means and SVR

为解决光伏功率预测不准确问题,提出了一种基于自适应k均值和支持向量回归的光伏出力预测方法。首先,分析了k均值聚类及其改进方法,给出了支持向量回归(SVR)的基本原理和应用流程,介绍了SVR中径向基函数凸优化模型。然后,结合自适应k均值和支持向量回归,依据光伏出力基本特点,分析了光伏出力预测流程及预测结果统计学评价指标。最后,以“云南昆明”光照数据为实际算例,确定了预测模型结构,并分别采用k-means and SVR、ARMA和ANN这3种方法进行预测,对比了不同聚类结果和不同算法时的预测统计指标,验证了所提方法的有效性,为光伏出力预测提供了一种方法。

To solve the problem of inaccurate photovoltaic power prediction,a photovoltaic output prediction method based on adaptive k-means and support vector regression is proposed.Firstly,k-means clustering and its improvement methods are analyzed,and the basic principle and application process of support vector regression(SVR) are presented.The radial basis function convex optimization model is introduced.Then,combining adaptive k-means and support vector regression,based on the basic characteristics of photovoltaic output,the photovoltaic output prediction process and statistical evaluation indicators of prediction results are analyzed.Finally,taking the lighting data of “Kunming,Yunnan” as an actual calculation example,the prediction model structure is determined,and three algorithms,k-means and SVR,ARMA,and ANN,are used for prediction.The prediction statistical indicators under different clustering results and algorithms are compared to verify the effectiveness of the proposed method,providing a method for photovoltaic output prediction.