基于改进LSTM的多工况机电设备剩余寿命预测

Residual Life Prediction of Electromechanical Equipment under Multiple Working Conditions via Improved LSTM

多工况设备在工况变化时存在特征值跳动问题,给剩余寿命预测带来挑战.本文提出了一种多工况设备剩余寿命预测方法.首先给出了多工况设备健康状态数据的标准化方法,降低工况变化对特征值的波动影响;其次提出了一种设备健康指标多源数据融合方法,将多维健康特征融合成一维健康指标,该指标可以反映设备的退化态势;再利用滑动窗口方法构建了训练和测试时间序列样本.为了降低训练样本不均衡导致的过拟合影响问题,提出了一种KNN样本权重赋值方法;最后建立了多层BLSTM剩余寿命预测网络,为了增强网络的预测精度,加入了改进的Huber损失函数.C-MAPSS数据实验结果表明本文提出的多工况设备剩余寿命预测方法具有很好的预测效果.

It is a challenge to predict the residual life of the electromechanical equipment under multiple working conditions because of the unstable characteristic value when the working condition changes.In this paper,a prediction method is proposed for the residual life of the equipment under multiple working conditions.Firstly,a standardized method of the health status data is proposed for the equipment under multiple working conditions,which can reduce the impact of working conditions on the characteristic value fluctuation.Secondly,a multi-source data fusion method is proposed for equipment health indicators,which can fuse multi-dimensional health features into one-dimensional health indicators,and the deterioration situation of the equipment can be reflected by the fused health indicators.Thirdly,the sliding window method is employed to construct the time series of the training and testing samples.A sample weight assignment method via the KNN is proposed to reduce the over-fitting problem caused by the unbalanced training samples.Finally,a multi-layer BLSTM prediction network with an improved Huber loss function is established to enhance the prediction accuracy of the network.The experimental results of the C-MAPSS data show a good prediction effect on the residual life of the equipment under multiple working conditions via the proposed prediction method.