叶片前缘造型对压气机平面叶栅气动特性的影响

为解决递归神经网络（RNN）模型难以训练和梯度消亡现象，引入长短期记忆网络算法( LSTM).介绍长短期记忆网络的基本原理，设计模型的更新算法，并将其应用于机械状态监测领域.以电机轴承数据为样本进行仿真，针对轴承数据的非平稳性，运用经验模态分解方法将其分解为平稳信号，计算本征模态分量能量熵，并将其作为状态特征.利用长短期记忆网络对机械状态单步预测结果与使用支持向量回归机模型的预测结果比较表明，长短期记忆网络在机械状态预测方面可以取得较支持向量回归机更好的效果.
In order to solve the train suffer and gradient extinction of recurrent neural network (RNN), a long-short term memory network (LSTM) algorithm was proposed and applied to the monitor of mechanical state (PMS) after introducing basic principle and designing updating algorithm of model. The simulation was carried out by taking the motor bearing data as samples, in view of the nonstationarity of bearing data, the empirical mode decomposition (EMD) was used to decompose the bearing data into stationary singles, and the intrinsic mode function (IMF) energy entropy was calculated as the feature of mechanical state. By comparing LSTM’s single-step prediction result of mechanical state with SVRM’s prediction result, it is shown that LSTM is better than support vector machine regression in terms of machine performance prediction.