基于时间卷积注意力网络的剩余寿命预测方法

Temporal convolutional attention network for remaining useful life estimation

剩余使用寿命(RUL)预测对于保障现代工业设备的安全运行、降低维护成本具有重要意义。目前已有基于循环神经网络(RNN)的RUL预测模型结构比较复杂,且缺乏一种从多传感器数据中提取重要退化信息的有效机制。因此,设计了一种新的用于RUL预测的时间卷积注意力网络(TCAN)模型。在TCAN中使用结构相对比较简单的时间卷积神经网络(TCN)来提取传感器数据中的退化特征,然后利用注意力机制从TCN中提取重要的退化特征信息。最后,将学习得到的高层特征表示展开并输入全连接层,输出预测的RUL值。在C-MAPSS数据集上与其他方法相比较,分析了TCAN模型的性能,实验结果表明TCAN可以更有效地提高RUL预测的精度。

Remaining useful life(RUL)prediction is of great significance for ensuring the safe operation of modern industrial equipment and reducing maintenance costs.At present,the existing RUL models based on recurrent neural networks are complex in structure and lack an effective mechanism to extract important degradation information from multi-sensor data.The new Temporal Convolutional Attention Network(TCAN)model was proposed for RUL estimation.A Temporal Convolutional Neural(TCN)network with a simple structure was used in TCAN to extract the degradation features from the sensor data,and then the attention mechanism was used to extract the important degradation information.The learned high-level feature representation was flattened and fed into a fully connected layer to output the predicted RUL.Compared with other methods on the C-MAPSS dataset,the experimental results showed that the TCAN could more effectively improve the accuracy of remaining life prediction.