基于DGA的粒子群极限学习机电力变压器故障诊断  

Fault Diagnosis of Power Transformer Using Particle Swarm Optimization and Extreme Learning Machine Based on DGA

作　　者：袁海满 吴广宁[1] 高波[1] 

机构地区：[1]西南交通大学电气工程学院,成都610031

出　　处：《高压电器》2016年第11期176-180,187,共6页High Voltage Apparatus

摘　　要：传统单一人工智能方法对变压器故障诊断中采用的大量不完备信息不能够有效处理,导致故障诊断准确率不高。为弥补这一不足,在全面分析粒子群算法(particle swarm optimization,PSO)和极限学习机(extreme learning machine,ELM)各自优势的基础上,构建了一种基于粒子群优化极限学习机的变压器故障诊断方法。该方法以DGA作为特征输入,利用粒子群算法对极限学习机的输入层权值和隐含层阈值进行优化,从而提高变压器故障诊断的精度。实例对比分析表明,与BP神经网络和极限学习机方法相比,粒子群极限学习(PSO-ELM)方法有更高的诊断准确率。Traditional single artificial intelligence method cannot effectively process the huge amount of incomplete fault information from transformer fault diagnosis resulting in low accuracy of fault diagnosis. In this paper, a novel transformer fault diagnogis method based on particle swarm optimization （PSO） and extreme learning machine（ELM） is proposed by making full use of the advantages of PSO and ELM. In this method,DGA is taken as the feature,PSO algorithm is adopted threshold, hence the accuracy of transformer fault to optimize ELM＇s input layer weights and implicit layer diagnosis can be improved. Examples contrast shows that, compared with the BP neural network and the traditional extreme learning machine, the proposed PSO-ELM method has higher diagnosis accuracy.

关 键 词：极限学习机 神经网络 粒子群算法 变压器 

分 类 号：TM41[电气工程—电器]