基于AdaBoost.M2-NN的变压器故障诊断

组合分类器的经典算法AdaBoost即自适应Boosting算法是提高预测学习系统预测能力的有效工具.针对传统BP（Back Propagation,BP）神经网络在变压器故障诊断时存在不稳定和网络易陷于极小值等缺点,将AdaBoost扩展算法AdaBoost.M2与BP神经网络结合,形成基于Ada-Boost.M2-NN（AdaBoost.M2Neural Network）的变压器故障诊断模型.利用AdaBoost的集成提升作用,在一定程度上弥补了BP算法的不足.仿真结果表明：该模型不仅能将单个BP神经网络无法识别的样本类别识别出来,而且还能整体上相比BP神经网络和传统三比值法将识别率提高11.5%,说明其具有可行性.

Fault Diagnosis of Transformer Based on AdaBoost.M2-NN

As a classical classifier combination method,AdaBoost is an effective tool to improve the predictive ability of predictive learning system.Because of the shortage of the conventional BP neural network when it is used in transformer fault diagnosis,the expansion type AdaBoost.M2 of AdaBoost algorithm is applied to fault diagnosis based on BP neural networks to set up an AdaBoost.M2-NN model for transformer fault diagnosis.The role of integration of AdaBoost makes up for the BP algorithm to some extent.AdaBoost.M2-NN model can not only identify the samples that the single BP neural network can not identify,but also raise the rate of identification by 11.5% as compared with the single BP neural network and the traditional three-ratio method.The results show that this method is reliable and feasible.