CEEMDAN and Informer Based Model for Aero-Engine Parameter Prediction

Predicting the parameter values of an aero-engine is beneficial to detect abnormal states in advance. Most existing methods directly predict the original parameters without adopting mode decomposition. This paper proposes a mode decomposition based fusion parameter prediction model, which can capture the transformation law of the raw signal and enjoy improved prediction accuracy. To this end, we utilize complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to decompose the raw time series data (N1r) to obtain multiple distinct mode components and residual signals. The complexity of each component is evaluated by fuzzy entropy, and the components with similar complexity are fused to reduce the input dimension of the prediction part. Informer is used to predict the fusion components, and the summation method is used to fuse each component to obtain the predicted value of N1r. Finally, the efficiency of the fusion model and the original Informer are verified over an engine’s flight log data. The results show that the proposed mode decomposition based fusion parameter prediction model has better performance than Informer.