CNCP:一种新型仿生的轻量级电能质量扰动信号分类模型

CNCP,Power Quality Disturbance Signal Classification Model Based on NCP

电能质量扰动信号分类对智能电网的稳定以及安全运行具有重要意义.针对当前电能质量信号分类模型中分类准确度不高、分类模型参数较多等问题,本文首次提出了一种基于NCP(Neural Circuit Policy)的轻量级电能质量信号分类模型CNCP(Classification-NCP),CNCP网络模型由信号特征提取、预分类和分类优化3个部分构成.首先,通过引入一维卷积核代替传统二维卷积核对信号进行特征提取,从而更有效的提取信号潜在特征.其次,通过优化NCP神经元数,减少了在特征处理过程中信息的损失.最后,通过全连接网络对预分类结果进行优化,提高了CNCP网络的泛化能力.在IEEE-1159电能质量扰动信号数据集上的实验结果表明,本文提出的CNCP与其他常用的电能质量扰动信号分类模型相比,分类结果准确性更高,抗噪性更强,模型参数更少.

The classification of power quality disturbances(PQD) is of great significance to the stability of the smart grid and the safety operation.Aiming at the problems of low accuracy and the redundant parameters in the current power quality disturbances classification model,this paper proposes a Neural Circuit Policy(NCP) based lightweight power quality disturbances classification model,namely classification-NCP(CNCP).The CNCP network model consists of three parts:signal feature extraction,pre-classification and classification optimization.Firstly,a one-dimensional convolution kernel is introduced to replace the traditional two-dimensional convolution check signal for feature extraction,so as to extract the potential features of the signal more effectively.Secondly,the number of neurons in NCP network is optimized to reduce the loss of information in the process of feature processing.Finally,the results of the pre-classification were optimized by the fully connected network to improve the generalization ability of the CNCP network.Experimental results on IEEE-1159 power quality disturbance signal dataset show that,compared with other commonly used power quality disturbance signal classification models,the proposed CNCP has higher classification accuracy,stronger noise resistance and fewer model parameters.