基于并行融合深度残差收缩网络的有源配电网故障诊断  

Research on fault diagnosis of active distribution network based on parallel fusion deep residual shrinkage network

作　　者：冯骥 杨国华[1,2] 史磊 潘欢 陆宇翔[1] 张元曦 李祯 FENG Ji;YANG Guohua;SHI Lei;PAN Huan;LU Yuxiang;ZHANG Yuanxi;LI Zhen(School of Electronic and Electrical Engineering,Ningxia University,Yinchuan 750021,China;Ningxia Key Laboratory of Electrical Energy Security,Yinchuan 750004,China)

机构地区：[1]宁夏大学电子与电气工程学院,银川750021 [2]宁夏电力能源安全重点试验室,银川750004

出　　处：《综合智慧能源》2024年第6期8-15,共8页Integrated Intelligent Energy

基　　金：国家自然科学基金项目(61763040);宁夏自然科学基金项目(2021AAC03062)。

摘　　要：针对含分布式电源的配电网故障呈现方式多样化以及故障诊断易受分布式电源类型、输出功率等非线性因素影响等问题,提出一种基于并行融合深度残差收缩网络(P-FDRSN)的故障诊断模型。首先,构建具有故障识别支路和故障定位支路的并行网络结构——P-FDRSN,在残差模块中引入收缩机制,减少网络中噪声或冗余信息的影响,提高网络对噪声的鲁棒性;其次,将故障录波信号波形幅值变化转换为灰度图和时频图,送入深度残差收缩网络进行深度特征提取并在汇聚层中将获取的特征进行融合,以增强故障录波信号的特征学习能力。仿真结果表明:在不同分布式电源类型和不同输出功率下,模型故障定位与识别精度均能保持在98.75%和97.25%以上,即使在噪声干扰的情况下,诊断准确率仍可保持在96.75%以上,模型具有较高的精度和较好的自适应性。Since the faults of the distribution networks with distributed generators present diversity,and the fault diagnoses are vulnerable to nonlinear factors such as the type of distributed generators and their outputs,a fault diagnosis model based on a parallel fusion deep residual shrinkage network(P-FDRSN)is proposed.The P-FDRSN is constructed by two parallel networks,a fault identification branch and a fault location branch.The parallel structure introduces shrinkage mechanism into its residual module to reduce the influence of noise or redundant information on the network and to improve the robustness of the network against noise.After transforming fault recording signal waveforms into grayscale images and time-frequency images,the signals are fed into the DRSN for deep feature extraction,and then,the acquired features are fused in the convergence layer,so as to enhance the feature learning capability on the fault recording signals.Finally,the simulation results show that the fault location and identification accuracies of the proposed model for various types of distributed generators of different outputs can be maintained above 98.75%and 97.25%,respectively.Even under the interference of noise,the diagnosis accuracy of the model will be kept above 96.75%,showing a high accuracy and decent robustness.

关 键 词：有源配电网 分布式电源 故障诊断 并行网络结构 并行融合深度残差收缩网络 

分 类 号：TM314[电气工程—电机]