基于TensorFlow框架的改进BP户变关系识别方法

An Improved BP Method for Transformer-User Relationship Identification Based on TensorFlow Framework

为了理清供电台区和用户之间的挂接关系,传统方法是依托人工或专用装置辨识,但存在效率低下和运行工况受限等问题。对此,在数据挖掘和机器学习技术基础之上,提出一种基于TensorFlow框架的改进BP(反向传播)户变关系识别方法。首先以Spearman相关系数作为K-means聚类的距离度量,对部分历史电压数据聚类,将聚类结果匹配所属台区进而构建训练样本,利用学习后的BP识别当前用户所对应的变压器及相别。为改善BP容易出现局部最优和收敛速度慢的问题,采用零均值化和Adam(自适应矩估计)对BP进行优化,并将BP部署在TensorFlow框架以进一步减小算法耗时。算例表明,所提算法能够有效提高户变关系识别准确率,提高辨识效率,具有良好的理论和应用价值。

To identify the connection relationship between transformers and users traditionally relies on manual work or special devices,which feature low efficiency and limited operating conditions.This paper,employing data mining and machine learning,presents an improved BP(backpropagation)method for transformer-user relationship identification based on the TensorFlow framework.Firstly,the Spearman correlation coefficient is used as the distance of K-means clustering.Some historical voltage data are clustered and the clustering results are matched with their transformers to construct training samples.Then the learned BP is used to identify the transformer and phase corresponding to the user voltage data.Meanwhile,to solve the problem of local optimization and slow convergence of BP,zero-mean and Adam(adaptive moment estimation),are used to optimize BP and deploy it in the TensorFlow framework to further reduce the time for computing.Examples show that the proposed algorithm can effectively improve the accuracy and efficiency of transformer-user relationship identification,and has good theoretical and application value.