基于改进BP神经网络的逆变电路故障诊断  

Fault diagnosis of inverter circuit based on improved bidirectional BP neural network

作　　者：廖俊勃 帕孜来.马合木提 蔡鑫[1] 周浩[1] 

机构地区：[1]新疆大学电气工程学院,新疆乌鲁木齐830047

出　　处：《电源技术》2015年第3期574-577,共4页Chinese Journal of Power Sources

基　　金：国家自然科学基金(61364010)

摘　　要：针对BP神经网络鲁棒性、容错性不强的问题,提出双向BP神经网络,更直接地建立与先前状态的映射关系;利用量子粒子群算法(QPSO)优化双向BP神经网络的权值和阈值,克服其学习算法复杂、收敛速度慢的缺点,来得到精度更高的网络。将改进的双向BP神经网络应用于逆变电路的故障诊断,测试结果表明该算法比双向BP神经网络具有更强的收敛性和精确率,为逆变电路的故障诊断提出一个新的思路。A kind of bidirectional BP neural network（BBPNN） was proposed aiming at BP neural network which robustness and fault tolerance was not strong. Reverse connection was established between adjacent layer neurons directly to draw the previous state variables of the system into the network. In order to overcome learning algorithm complexity and slow convergence, quantum particle swarm optimization（QPSO） was used to optimize weights and thresholds of BP neural network. In this way, the network could be obtained with higher accuracy. Improved bidirectional BP neural network was applied to the inverter circuit fault diagnosis. Test results show that the algorithm has stronger convergence and accuracy than BBNPP. A new way for the inverter circuit fault diagnosi was proposed.

关 键 词：双向BP神经网络 量子粒子群算法 逆变电路 故障诊断 

分 类 号：TM464[电气工程—电器]