基于概率稀疏自注意力的航空发动机剩余寿命预测

航空发动机剩余寿命预测对其健康管理具有重要意义,针对长序列、多维度的航空发动机监测参数,提出一种基于概率稀疏自注意力（ProbSparse Self-Attention）的Transformer模型以实现航空发动机剩余寿命的准确预测。用ProbSparse Self-Attention取代原始Transformer中的常规自注意力机制,使得模型更关注时间序列中重要的时间节点,大幅缩减时间维度,减小了时间和空间复杂度;通过注意力层整合后的信息,进一步通过前馈神经网络层和卷积层,提取传感器的空间特征,编码层之间通过扩张因果卷积相连接,扩大了感受野,提高了模型对长序列信息的捕获能力。在新公开的N-CMAPSS数据集上验证算法,实验结果表明,相比于实验中的对比模型,所提模型的RMSE和Score值均有提升,推理速度也优于其他模型。

Remaining Useful Life Prediction of Aero-engine Based on ProbSparse Self-Attention

The remaining life prediction of aero-engine was of great significance for its health management. Aiming at the long sequence and multiple dimensions aero-engine monitoring parameters, a Transformer model based on probabilistic sparse self-attention was proposed to realize the accurate prediction of the remaining life of the aero-engine. The regular Self-Attention mechanism in the original Transformer was replaced by ProbSparse self-attention, which made the model pay more attention to the important time nodes in the time series, greatly reduces the time dimension, and reduced the time and space complexity. Through the integrated information of the attention layer, the spatial features of the sensor were further extracted through the feedforward neural network layer and the convolutional layer. The encoding layer was connected by the dilated causal convolution, which expanded the receptive field and improves the model"s ability to capture long sequence information. The algorithm was verified on the newly public N-CMAPSS dataset. The experimental results show that compared with the comparison models in the experiment, the RMSE and Score values of the proposed model are improved, and the inference speed is also better than other models.