Remaining Useful Life Estimation of Aeroengine Based on Multi-head Attention LSTM Model and Genetic Algorithm

Accurate remaining useful life (RUL) prediction is very necessary for the aeroengine. In this paper, a novel joint prediction model based on multi-head attention LSTM and genetic algorithm (MHALN-GA) is proposed to address the two issues, which are insufficient extraction of multidimensional data features of engines using existing methods and insufficient guarantee of optimal model output solutions. Firstly, the operation of embedding multi-head attention modules into LSTM cells enables the attention mechanism to naturally participate in the model prediction process and increase the flexibility of model. Secondly, the addition of genetic algorithms avoids the difficulty of determining whether the model results are optimal, greatly improving the accuracy of model prediction while saving computational time. To verify the effectiveness of the proposed method, experiments were conducted on the C-MAPSS dataset, and the results showed that compared with SOTA methods, the MHALN-GA model has more accurate predictions and smaller errors, RMSE and Score have decreased by 15.51% and 12.26% respectively on FD001, and also have shown good performance on FD004.