基于改进集成聚类和BP神经网络的电压偏差预测

Voltage deviation forecasting based on improved ensemble clustering and BP neural network

电能质量问题日益严重,其中电压偏差的危害最为明显。本文提出一种基于改进集成聚类和BP神经网络的电压偏差预测方法。针对单一聚类算法的不足,将AP聚类算法与经典K-means聚类算法结合,形成改进集成聚类算法,实现两类算法的优势互补,该算法包括PCA降维、AP聚类、K-means聚类三步。选择与待预测点相似相近的样本数据集,采用改进集成聚类算法对数据集中的气象数据进行聚类,提取训练样本,最后采用BP神经网络算法建立预测模型。结果表明,该方法预测结果平均相对误差为2.987%,优于传统BP神经网络预测模型以及结合PCA降维的BP神经网络预测模型。

This paper proposes a voltage deviation forecasting method based on improved ensemble clustering andBP neural network. The ensemble algorithm was a combination of principal component analysis （ PCA）, affinitypropagation clustering and K-means clustering. Firstly the PCA algorithm was used for lowering the dimensions ofthe meteorological data, and then the improved ensemble cluster analysis is performed on the principal components,and finally the BP neural network for voltage deviation forecasting is adopted. The practical application results provethat the eMAPE of the proposed method is 2. 987%, which is obviously better than that of the traditional BP modeland BP model with PCA.