基于可自动扩展的LSTM模型的航空发动机剩余寿命预测方法

对航空发动机进行实时状态监测和健康管理可以有效降低发动机故障风险，确保飞机飞行安全。准确预测航空发动机的剩余寿命是有效监测发动机运行状态的一种重要手段，其中长短期记忆（long-short term memory，LSTM）网络常被使用。但由于航空发动机复杂的机械结构与运行模式，使用传统的LSTM网络对航空发动机的剩余寿命进行单次预测后，所得预测结果的准确率不足以满足其寿命预测的精度要求。基于LSTM网络的广泛使用以及它对时间序列数据的有效预测能力，考虑到采用多级预测的方法能够有效降低预测误差，提出了一种新型的可自动扩展的长短期记忆（automatically expandable LSTM，AELSTM）预测模型。AELSTM模型依托多个子模块逐级连接的网络结构，不断地提取前一级模块的输出误差作为后一级模块的训练值，形成了误差的多级预测机制，有效降低了模型的预测误差，提升了预测结果的准确性。基于美国国家航空航天局发布的C-MAPSS数据集的四个子集对AELSTM模型的预测效果进行了测试，实验结果表明，与传统的LSTM网络相比，AELSTM模型在四个子集上的均方根误差平均减少了95.44%，同时它的预测效果也优于现有的一些先进算法。实验充分验证了AELSTM模型在提升航空发动机剩余寿命预测准确度方面的有效性及优势。

Remaining useful life prediction method of aero-engine based on auto-expandable LSTM model

Real-time condition monitoring and health management of aero-engines can effectively reduce the risk of engine failure and ensure the safety of aircraft flight. Accurate prediction of the remaining life of an aero-engine is an important tool for effective monitoring of the engine operating condition, in which long-short term memory(LSTM) networks are often employed. However, because of the complex mechanical structure and operation mode of aero-engine, the accuracy of the prediction results obtained after a single prediction of the remaining life of the aero-engine using the traditional LSTM network is not sufficient to meet the accuracy requirements of its life prediction. Based on the widespread use of LSTM networks and their ability to effectively predict time series data, and considering that a multi-level prediction approach can effectively reduce the prediction error, this paper proposed a novel automatically expandable LSTM(AELSTM) prediction model. The AELSTM model relied on the network structure of multiple sub-modules connected level by level, and continuously extracted the output error of the previous level module as the training value of the next level module, forming a multi-level prediction mechanism of the error, which effectively reduced the prediction error of the model and improved the accuracy of the prediction results. Finally, this paper tested the predictive effectiveness of the AELSTM model based on four subsets of the C-MAPSS dataset published by NASA. The experimental results indicate that the root-mean-square error of the AELSTM model decreases by 95.44% on average over the four subsets compared with the traditional LSTM network, and it also outperforms some existing state-of-the-art algorithms in prediction. The experiments fully verify the effectiveness and advantages of the AELSTM model in improving the accuracy of remaining life prediction of aero-engines.