基于人工免疫分类算法的电力变压器故障诊断

Power Transformer Fault Diagnosis by Using the Artificial Immune Classification Algorithm

基于人工免疫识别原理提出一种电力变压器故障诊断方法。该算法模拟自然免疫中抗原和B细胞相互作用机制,将故障样本(变压器油中溶解气体体积)作为抗原,用加权欧氏距离计算亲和力,兼顾分量单项超标故障信息,通过免疫训练,获取表征故障样本的人工识别球集合,再用最邻近分类法对故障样本分类。实例表明,该算法能有效识别变压器故障,具有较高的检测准确率,在电力变压器故障诊断中有良好的应用前景。

An artificial immune recognition algorithm was proposed for fault diagnosis. The interaction mechanism between antigens and B ceils was simulated, where fault sample （feature vector consists of ingredients dissolved gasin-oil in transformers） was regarded as antigen. The principle was that, the highest affinity B ceils cloned and mutated for diversification and shapespace exploration, then developed into ARB（artificial recognition ball） , at the same time, class information was added to ARB so that it was trained to learn the feature of fault samples better. Here, affinity was computed with weighted Euclidean distance in order to take some single abnormal ingredient into account. Subsequently, antigens were presented to ARBs, and ARBs competed for representation more B cell until stimulation value of each ARB reached the threshold. In this way, the ARB pool was generated and served as a classification tool. After training being completed, the classification was performed in a k-nearest neighbor approach. The classification of sample was determined by using a majority vote of the outputs of the five most stimulated ARBs, and a mass of fault samples were tested. The results indicate that the presented algorithm has potent classification capability as well as high diagnosis precision, and this method has promising application potential in power transformer fault diagnosis.