基于SCG优化SSAE-FFNN的电能质量复合扰动深度特征提取与分类  

Power quality composite disturbance deep feature extraction and classification based on SCG optimized SSAE-FFNN

作　　者：丁皓月 吕干云[1] 史明明 费骏韬 俞明 吴启宇 DING Haoyue;LYU Ganyun;SHI Mingming;FEI Juntao;YU Ming;WU Qiyu(School of Electric Power Engineering,Nanjing Institute of Technology,Nanjing 211167,China;State Grid Jiangsu Electric Power Co.,Ltd.Research Institute,Nanjing 211103,China;State Grid Nanjing Lishui Power Supply Company of Jiangsu Electric Power Co.,Ltd.,Nanjing 211200,China)

机构地区：[1]南京工程学院电力工程学院,江苏南京211167 [2]国网江苏省电力有限公司电力科学研究院,江苏南京211103 [3]国网江苏省电力有限公司南京市溧水区供电分公司,江苏南京211200

出　　处：《电力工程技术》2024年第3期99-110,共12页Electric Power Engineering Technology

基　　金：江苏省高等学校自然科学研究重大项目“含高渗透率分布电源配电网电压暂降鲁棒状态估计及应用”(19KJA510012);江苏省南京工程学院校基金(YKJ202209)资助。

摘　　要：随着智能电网的发展,电能质量问题已遍布电网并威胁着电网的安全稳定,且电能质量监测数据日渐庞大,因此实现大规模系统中电能质量扰动(power quality disturbances,PQDs)的深度特征提取及智能分类识别对电力系统污染检测与管理具有重要意义。为此,文中提出一种基于堆叠稀疏自编码器(stacked sparse auto encoder,SSAE)和前馈神经网络(feedforward neural network,FFNN)的电能质量复合扰动分类方法。首先,基于IEEE标准构建PQDs仿真模型。然后,建立基于SSAE-FFNN的PQDs分类模型,并引入缩放共轭梯度(scaled conjugate gradient,SCG)算法对模型进行优化,以提高梯度下降速度和网络训练效率。接着,为有效降低堆叠网络的重构损失同时提取出深度的低维特征,构建SSAE的逐层训练集及微调策略。最后,通过算例分析验证文中方法的分类效果、鲁棒性、泛化性和适用场景规模。结果表明,文中方法能够有效识别电能质量复合扰动,对含误差扰动和某地市电网的21组实测扰动录波数据也有较高的分类准确率。With the development of the smart grid,power quality issues have been widespread in the power grid and it threaten the safety and stability of the power grid.The monitoring data of power quality disturbances(PQDs) increase rapidly,and it is of great significance to achieve deep feature extraction and intelligent classification of PQDs in large-scale systems for power system pollution detection and management.To this end,stacked sparse auto encoder(SSAE) and feedforward neural network(FFNN) based method for composite PQDs classification is proposed in this paper.Firstly,a PQDs simulation model is constructed based on IEEE standard.Then,a PQDs classification model based on SSAE-FFNN is established,and the scaled conjugate gradient(SCG) algorithm is used to optimize the model,in order to accelerate gradient descent and improve training efficiency.Next,to reduce the reconstruction loss of the stacked network and extract deep low-dimensional features,the layer-wise training and fine-tuning strategy of SSAE are constructed.Finally,the examples are used to verify the classification effect,robustness,generalization and applicable scenario scale of the proposed method.The results show that the method can effectively identify composite PQDs and it has a high accuracy even for both error-containing disturbances and 21 sets of measured disturbance data of a local municipal grid.

关 键 词：电能质量 复合扰动分类 堆叠稀疏自编码器(SSAE) 深度特征提取 缩放共轭梯度(SCG) 前馈神经网络(FFNN) 

分 类 号：TM732[电气工程—电力系统及自动化]