基于E2E DeepVAE-LSTM的轴承退化预测应用研究

针对额外提取数据特征的方法需要花费大量时间和人力成本，轴承退化的线性预测精度低等问题，以及时序数据具有时间依赖关系的特点，提出了端到端的结合长短时记忆网络的深度变分自编码器模型（E2E Deep VAE-LSTM）用于轴承退化预测。通过改进VAE的结构，并结合LSTM，该模型可以在含有异常值的数据集上直接进行训练和预测；使用系统重建误差表征轴承退化趋势，实现了轴承退化的非线性预测。在三个真实数据集上的实验结果表明，E2E Deep VAE-LSTM模型可以得到满意的预测结果，预测精度均高于现有的几种AE类模型及其他几种方法，且具有良好的泛化能力和抗过拟合能力。

Application research on bearing degradation prediction based on E2E Deep VAE-LSTM

Aiming at the problems that the methods of additionally extracting data features needed a lot of time and labor cost, the low accuracy of linear prediction of bearing degradation, and the time dependence of time series data, this paper proposed an end-to-end deep variational autoencoder(VAE) model combined with long-short term memory networks(E2E Deep VAE-LSTM) for bearing degradation prediction. By improving the structure of VAE and combining with LSTM, this model directly trained and predicted on datasets containing outliers, and used the system reconstruction error to characterize the bearing degradation trend, realized the nonlinear prediction of bearing degradation. Experimental results on three real datasets show that the E2E Deep VAE-LSTM model can obtain satisfactory prediction results, the prediction accuracy is higher than several existing AE models and other methods, and it has good generalization ability and anti-overfitting ability.