基于稀疏自动编码器神经网络的负荷曲线分类方法

Power Load Profile Classification Method Based on Neural Network of Sparse Automatic Encoder

随着电力市场精细化发展以及电力大数据的广泛应用,深度探索电力用户用电行为特性具有重要意义,因此该文提出一种结合有监督和无监督算法的电力负荷曲线分类方法。首先,基于距离与曲线形态的双尺度相似性度量,采用无监督优化谱聚类算法获得负荷曲线精准标签数据;其次,采用稀疏自动编码器神经网络学习大规模待分类负荷曲线的内在特征,得到隐藏层权值矩阵即神经网络的优化初始参数;最后,基于已获得的标签数据,训练支持向量机神经网络分类器,实现对大规模待分类负荷曲线的有监督分类。基于爱尔兰负荷数据,算例表明本文提出的分类方法在DBI指数、轮廓系数、分类有效性以及计算速度等方面具有更好的性能。

With the elaborate development of the electricity market and the wide applications of power electricity big data,it is of great significance to deeply explore the behavior characteristics of the power users.In this paper,a method combining supervised algorithm and unsupervised algorithm is proposed for the power load profile classification.Firstly,the accurate profile label data is obtained by using the unsupervised optimized spectral clustering algorithm based on the two-scale similarity measures of the distance and the shape.Secondly,the intrinsic features of the large-scale load profiles are recognized through adopting the sparse auto-encoder neural network to obtain the weighting factor matrix of the hidden layer,i.e.the optimal initial parameter of the neural network.Finally,based on the obtained tag data,the neural network classifier supporting the vector machine is used to realize the supervised classification of the large-scale load profiles.With the Irish CER electricity data,the case study proves that the proposed classification method has a better performance on the DBI index,the Silhouette coefficient,the classification validity and the calculation speed.