摘要
针对传统算法在变压器故障诊断领域存在参数难以选取、准确率低、易误判等缺点,提出一种基于改进深度信念网络(IDBN)的电力变压器故障诊断方法。在油中溶解气体分析(DGA)基础上,首先以IDBN无监督训练方式重构原始数据特征,然后以有监督方式学习特征与故障类型之间的映射关系,最后将测试数据应用于模型并进行实验。实验结果表明,该方法不仅具有较高精度,而且在准确率方面优于传统的人工神经网络和支持向量机方法。因此,将改进深度信念网络用于变压器故障诊断具有较高的应用价值。
Aiming at the shortcomings of traditional algorithms including difficulties to select the parameters in the field of transformer fault diagnosis accuracy and low false judgment,we propose an improved fault diagnosis method of power transformer based on improved depth belief network（IDBN）.On the basis of dissolved gas analysis（DGA）in oil,we reconstruct the original data characteristics by unsupervised training method of IDBN,and then study the mapping relationship between features and fault types in a supervised manner.Finally,the test data is applied on the model and the experiment is carried out.The results show that the proposed method not only has higher precision,but also outperforms the traditional artificial neural network and support vector machines in accuracy.Therefore,it is of practical value and practical significance to improve the deep belief network for transformer fault diagnosis.
作者
李辉
张志攀
LI Hui;ZHANG Zhi-pan(School of Physics and Electronic Information,Henan Polytechnic University;School of Electrical Engineering ancl Automation,Henan Polytechnic University,Jiaozuo454000,China)
出处
《软件导刊》
2018年第7期169-172,共4页
Software Guide
基金
河南基础与前沿技术研究计划项目(152300410103)
河南省教育厅科学技术研究重点项目(13A510330)
关键词
改进深度信念网络
变压器故障
气体分析
准确率
improved deep belief network
transformer fault
gas analysis
accuracy rate
