一种基于生成对抗域的长短期记忆网络系统剩余寿命预测方法

A Remaining Useful Life Prediction Method for Long-short Term Memory Network Systems Based on Generative Adversarial somains

针对系统监测数据存在不平衡而导致剩余寿命预测不准确的问题,本文提出一种基于生成对抗域的长短期记忆网络系统剩余寿命预测方法。首先,提取原始振动信号,构建训练数据集和测试数据集。其次,采用长短期记忆网络替换Wasserstein生成对抗网络中的生成器模块,构建一个新的Wasserstein生成对抗网络框架并扩增原始数据,建立平衡数据集。最后,将数据集送入长短期记忆网络进行剩余寿命预测,并通过PHM2012公开的轴承退化数据集和航空发动机数据集进行验证。实验结果表明,Wasserstein生成对抗网络可以改善数据不平衡状态,并且利用平衡数据集训练长短期记忆网络,能够有效提高性能退化趋势预测能力和预测精确度。

To solve the problem of inaccurate remaining useful life prediction caused by the imbalance of bearing monitoring data,this paper proposes a RUL prediction method based on the generation of long-short term memory network in the confrontation domain.At first,the vibration signals of the original data are extracted to construct the training dataset and test dataset.Next,a long-short term memory network is used to replace the generator module in the Wasserstein generative adversarial network to construct a new Wasserstein generative adversarial network framework and augment the original data to create a new equilibrium dataset.Finally,the dataset is fed into the long-short term memory network for remaining useful life prediction and the performance of the proposed method is evaluated by the bearing degradation dataset and the aero-engine dataset disclosed in PHM2012.The experimental results show that Wasserstein can improve the data imbalance state by generating the confrontation network.Using the balanced data set to train the long-short term memory network can effectively improve the prediction ability of the performance degradation trend and improve the prediction accuracy.