基于GRU Encoder-decoder和注意力机制的RUL预测方法

Method of RUL prediction based on GRU Encoder⁃decoder and attentional mechanism

深度学习模型可直接建立机械设备的状态与剩余使用寿命(RUL)之间的映射关系,从而避免人工提取特征和建立健康指标的过程。文中基于深度学习理论,提出一种基于注意力机制和时序编码解码器(Encoder-decoder)相结合的RUL预测方法。首先,基于门控循环神经网络(GRU)构建一个时序编码解码器以实现输入序列的重构,其中GRU-Encoder对输入的多元时间序列进行编码;再引入注意力机制对GRU-Encoder在每个时刻的输出向量进行加权融合,以融合后的向量作为编码结果,并将其输入到GRU-Decoder中实现输入序列的重构,同时将编码结果映射为输入样本的RUL。采用CMAPSS数据集对所提方法的有效性进行验证,结果表明,该方法预测精度较高,可行且有效。

The deep learning model can directly establish the mapping relationship between the state of mechanical equipment and the remaining service life(RUL),thus avoiding the process of manually extracting features and establishing health indicators.Based on the deep learning theory,a RUL prediction method based on the combination of attention mechanism and Encoder⁃decoder is proposed.Based on the gated recurrent neural network(GRU),a timing codec is built to reconstruct the input sequence,in which GRU⁃Encoder can encode the input multivariate time series.The attention mechanism is introduced to weight and fuse the output vectors of GRU⁃Encoder at each time.The fused vector is used as the coding result,and input into GRU⁃Decoder to reconstruct the input sequence.At the same time,the coding result is mapped to the RUL of the input sample.The C⁃MAPSS data set is used to verify the effectiveness of the proposed method,and the results show that the prediction accuracy of the method is high,feasible and effective.