基于格拉姆角场和改进残差网络的低压配电台区户变关系识别方法

Consumer-Transformer Relationship Identification Based on GAF and Improved Residual Network Algorithm

低压配电台区户变关系信息是电力网络的基础性档案,准确识别低压配电台区户变关系对于电网企业尤为重要。采用一维序列分类神经网络进行台区户变关系识别时,存在序列易丢失时间依赖性、识别准确率低、准确率不稳定的问题,为此,提出一种基于格拉姆角场和改进残差神经网络的户变关系识别方法,以用户电压数据为基础,首先采用格拉姆角场方法将一维电压数据序列转换为特征矩阵,将矩阵元素对应于图像灰度采用伪彩色处理形成二维特征图谱;然后采用引入空间注意力模块进行改进的残差神经网络分类特征图谱以识别台区户变关系,采用格拉姆角场和伪彩色处理生成特征图谱,同时保留序列时间依赖性并使用空间注意力模块突出台区电压特征图谱差异,提高户变关系的识别准确率和稳定性。经实验验证,所建立的户变关系识别(CTRI)模型中台区用户识别平均准确率为98.52%,相较于一维卷积神经网络(1D-CNN)和二维卷积神经网络(2D-CNN)的平均识别准确率分别提升18.26%和10.1%。

Consumer-Transformer relationship is the basic file of the power network,and the ambiguity of the relationship between consumer and transformer will affect many tasks such as intelligent fault location,fault repair,and power outage information release of power grid companies and cause unnecessary manpower and material consumption.For this reason,it is particularly important for grid companies to accurately identify the relationship between consumer and transformer in low-voltage distribution stations.At present,the identification methods of consumer and transformer relationship are mainly divided into three categories:data dimensionality reduction and clustering,feature distortion and comparison,and big data mining.Big data mining to recognize the relationship between consumer and transformer mostly uses a one-dimensional sequence classification neural network,but there is a problem of time-dependent sequence of easily lost,low accuracy recognition,recognition accuracy unstable.Aiming at these problems,a method of consumer-transformer relationship identification based on Gram angle field(GAF)and improved residual neural network using consumer’s voltage data sequence in low-voltage station area is proposed.First,the Gram angle field method is used to convert the one-dimensional voltage data sequence into a feature matrix,the matrix elements are corresponded to the image gray level,and pseudo-color processing is used to form a two-dimensional feature map.Then the improved residual neural network with the introduction of spatial attention module is used to identify the relationship between consumer and transformer.The recognition method uses the Gram angle field and pseudo-color processing to generate a feature map.At the same time,the time dependence of the sequence is retained,and the spatial attention module is used to highlight the difference of voltage characteristic spectrum in the station area,so as to improve the recognition accuracy and stability of household transformer relationship.The experimental results show that the average accuracy of station area user recognition in the established user change relationship recognition(CTRI)model is 98.52%.Compared with the one-dimensional convolutional neural network(1D-CNN)and the two-dimensional convolutional neural network(2D-CNN),the average recognition accuracy is improved by 18.26%and 10.1%,respectively.