基于LSTM-SVM的风电机组齿轮箱故障诊断

Fault Diagnosis of Wind Turbine Gearbox Based on LSTM-SVM

针对风电机组齿轮箱的故障诊断中特征提取过分依赖人为经验和准确率不高的问题,提出一种基于长短时记忆网络(LSTM)与支持向量机(SVM)相结合的方法。对原始时域振动信号作傅里叶变换,利用LSTM神经网络自适应智能提取特征的优势,结合SVM的分类功能,实现对风电机组齿轮箱更加准确的故障诊断。仿真结果显示,该网络模型在经过16轮训练后准确率可以达到100%,使用测试集数据准确率也可以达到99.1%。

Aiming at the problem that feature extraction relies too much on human experience in the fault diagnosis of wind turbine gearboxes and the accuracy is not high,a method based on the combination of long short-term memory(LSTM)and support vector machine(SVM)was proposed.The FFT transformation was done on the original time-domain vibration signal,and the advantages of adaptive intelligent extraction of features by LSTM neural network were used,combined with the classification function of SVM,to achieve more accurate fault diagnosis of wind turbine gearbox.The simulation results show that the accuracy of the network model can reach 100%after 16 rounds of training,and the accuracy using test set data can also reach 99.1%.