基于改进粒子群优化的卷积神经网络在变压器故障诊断中的应用  

Application of CNN Based on Improved PSO in Transformer Fault Diagnosis

作　　者：种俊龙 郑莉 庄先涛 ZHONG Junlong;ZHENG Li;ZHUANG Xiantao(Suining Power Supply Company,State Grid Sichuan Electric Power Company,Suining,Sichuan 629000,China)

机构地区：[1]国网四川省电力公司遂宁供电公司,四川遂宁629000

出　　处：《自动化应用》2024年第20期105-107,共3页Automation Application

摘　　要：提出了一种基于改进粒子群(PSO)优化的卷积神经网络(CNN)方法,用于变压器的故障诊断。该方法将振动信号转化为时频图像,利用CNN自动提取特征并进行故障分类。为了提高诊断性能,采用改进PSO算法优化网络模型的超参数,并引入自适应权重策略和相对基学习策略增强算法的全局搜索能力。结果表明,所提方法能够有效提高变压器故障诊断的准确性,推动电力系统向智能化、自动化的方向发展。This paper presents a convolutional neural network(CNN)method based on improved particle swarm optimization(PSO)for fault diagnosis of transformers.The method transforms vibration signals into time-frequency images,and utilizes CNN to automatically extract features and classify faults.To enhance diagnostic performance,an improved Particle Swarm Optimization algorithm is employed to optimize the hyperparameters of the CNN,introducing an adaptive weight strategy and a relative base learning strategy to enhance the algorithm's global search capability.The results demonstrate that the proposed method can effectively improve the accuracy of transformer fault diagnosis,promoting the development of power systems towards intelligence and automation.

关 键 词：粒子群 变压器 故障诊断 卷积神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]