基于深度学习的电能质量扰动分类辨识  

Deep Learning-Based Classification and Identification of Power Quality Disturbances

作　　者：袁于程 黄健 谢晨旸 YUAN Yucheng;HUANG Jian;XIE Chenyang(Jiangsu Linyang Energy Co.,Ltd.,Qidong 226200,China)

机构地区：[1]江苏林洋能源股份有限公司,江苏启东226200

出　　处：《自动化仪表》2023年第3期68-73,共6页Process Automation Instrumentation

摘　　要：随着我国电力事业的飞速发展,人们对电能质量的关注度越来越高。为满足用户对电能质量的高要求,电能质量的治理尤其重要。电能质量扰动的分类辨识是电能质量治理的前提。提出了一种电能质量扰动的分类辨识算法。该算法只需输入原始扰动信号图像,便可辨识单一和叠加扰动信号的具体类别。首先,在数学建模的基础上,使用Matlab仿真单一和叠加扰动信号随时间变化的图像。其次,在特定频率下进行扰动信号采样,通过相空间重构法将采样获得的一维数据转换为二维平面轨迹图。再次,基于指定训练样本空间和标签,构建所需的卷积神经网络结构并经过迭代更新确定最终的网络结构训练参数。最后,对各类单一和叠加扰动进行分类辨识正确率的统计。仿真结果表明,该算法在不同信噪比的测试条件下具有良好的辨识正确率。仿真结果验证了该算法具有较强的抗噪性和鲁棒性。With the rapid development of China’s electric power industry,people pay more and more attention to the power quality.To meet the high requirements of users for power quality,power quality management is especially important.Identification of power quality disturbances is the premise of power quality management.A classification and identification algorithm for power quality disturbances is proposed.The algorithm can identify the specific classes of single and superimposed disturbance signals by simply inputting the original disturbance signal image.Firstly,based on mathematical modeling,the images of single and superimposed disturbance signals over time are simulated by Matlab.Secondly,the perturbed signals are sampled at specific frequencies,and the sampled one-dimensional data are converted into two-dimensional planar trajectory maps by using the phase space reconstruction method.Then,based on the specified training sample space and labels,the desired convolutional neural network structure is constructed and iteratively updated to determine the final network structure training parameters.Finally,the statistics of the correct classification recognition rate are performed for various types of single and superimposed perturbations.The simulation results show that the algorithm has good recognition correct rate under different signal-to-noise ratio test conditions.The results verify that the algorithm has strong noise immunity and robustness.

关 键 词：电网质量 扰动信号 相空间重构 二维平面轨迹图 卷积神经网络 训练参数 抗噪性 鲁棒性 

分 类 号：TH-89[机械工程]