基于自组织抗体网络的电力变压器故障诊断

Fault Diagnosis for Power Transformer Based on the Self-Organization Antibody Net

仿生生物免疫系统中抗体对抗原的高效识别和记忆机理,本文提出自组织抗体网络和抗体生成算法用于解决电力变压器故障诊断问题。自组织抗体网络中抗体类型与抗体浓度新定义的设计,优化了网络性能,只需根据样本数据设置初始抗体个数,无需人工设置任何其他参数与阈值。抗体生成算法依据抗体的类型和浓度,针对不同情况采取抗体进化、抗体合并以及抗体新生三种不同的策略,快速提取和记忆抗原特征,有效地提高了算法的效率。UCI(University of California,Irvine)标准数据集和基于油中溶解气体数据的电力变压器故障诊断试验表明,该方法能充分利用先验信息,实施有效的分类,并有很高的准确率。

Inspired by the highly efficient antibody-antigen recognition and memory mechanism of biological immune system,a self-organization antibody net（soAbNet） and the antibody generation algorithm are proposed and applied to fault diagnosis for power transformer.By the new definitions of antibody style and antibody density,the soAbNet works well with the initial number of antibodies,need not set any other artificial parameters and thresholds.According the antibody generation algorithm,antibodies learn and memory the characters of antigens effectively by three different strategies：antibody evolution,antibody combination and antibody production.Experimental result on Iris dataset from the UCI and diagnosis result on dissolved gas analysis data demonstrate that the proposed soAbNet can make full use of a priori information,it has effective classifying capability as well as higher precision.