摘要
针对感应电动机故障征兆与故障模式之间的复杂性和实际系统中的非线性给故障诊断带来的困难,采用一种把放大网络梯度函数(MGF)和附加动量项的自适应学习速率(ABPM)算法相结合的混合型方法(MABPM)建立感应电动机的神经网络故障诊断模型.通过与附加动量项的标准BP算法、ABPM算法、Polak-Ribiere共轭梯度算法和RPROP算法相比较,表明了MABPM算法具有更好的泛化稳定性和全局收敛性,故障诊断的平均准确率高于其他算法,并具有良好的诊断效果.
Considering the difficulty in fault diagnosis due to the complexity between the fault symptom and fauh pattern of induction motor and nonlinearity in actual system, a hybrid training approach （ MABPM）, combined with magnified network gradient function （MGF） and adaptive learning rate backpropagation with momentum （ABPM）, is adopted to construct the fault diagnosis model of induction motor based on neural network. When compared with the algorithms of standard backpropagation with momentum, ABPM, Polak-Ribiere conjunction gradient and RPROP, MABPM has better abilities of stable generalization and global convergence. The average fault diagnosis accuracy is enhanced compared with other algorithms, which shows promising diagnosis effectiveness.
出处
《天津理工大学学报》
2008年第1期7-10,共4页
Journal of Tianjin University of Technology
基金
天津市自然科学基金(2006DFA12410)
