摘要
提出了一种基于粒子群优化BP神经网络风电机组齿轮箱故障诊断方法。粒子群算法不需要计算梯度,可以兼顾全局寻优和局部寻优。利用粒子群算法对BP网络权值和偏置进行优化,减少了BP神经网络算法陷入局部最优解的风险,提高了神经网络的训练效率,加快了网络的收敛速度。考虑风电齿轮箱振动信号的不确定性、非平稳性和复杂性,提取功率谱熵、小波熵、峭度、偏度、关联维数和盒维数作为故障特征。经测试,算法诊断结果正确,表明了PSO优化BP神经网络用于风电机组齿轮箱故障诊断的有效性和实用性。
A method based on BP neural networks trained by particle swarm optimization (PSO) algorithm was presented for fault diagnosis of wind turbine gearbox. In the PSO algorithm, there is no need to calculate gradient, with consideration to both global optimization and local optimization. It can reduce the risk of BP neural network al- gorithm falling into local minimum, improve the training efficiency, and speed up convergence by using PSO algo- rithm to optimize the weights and bias of BP neural network. Power spectral entropy, wavelet entropy, kurtosis, skewness, correlation dimension and box dimension were extracted as fault feature for gearbox of wind turbine con- sidering uncertainty, nonstationarity and complexity of vibration signal. The method was tested and results of fault diagnosis are right. The validity and practicability of BP neural network algorithm trained by PSO algorithm for the wind turbine gearbox fault diagnosis were proved.
出处
《太阳能学报》
EI
CAS
CSCD
北大核心
2012年第1期120-125,共6页
Acta Energiae Solaris Sinica
基金
国家高技术研究发展(863)计划项目(2007AA05Z428)