Novel Extreme Learning Machine Using Kalman Filter for Performance Prediction of Aircraft Engine in Dynamic Behavior

Abstract

In this paper, a novel state-propagation extreme learning machine using a Kalman filter (KF-ELM) is proposed. In comparison with the plain extreme learning machine, the proposed algorithm takes the topological parameters as state variables and minimizes the covariance of state estimates to overcome the state conjunction and transformation dilemma in time series. As a result, its topological stability and prediction accuracy are enhanced, and these merits are further proved theoretically. In addition, the computational effort of KF-ELM is on the same order of magnitude as the plain extreme learning machine, while the former possesses a faster convergent speed. Then, several benchmark datasets are utilized to test the effectiveness and soundness of the proposed algorithm. Finally, it is employed to predict the gas path performance of a turbofan engine. The performance prediction accuracy is better than the plain ELM with different input rules in the dynamic process. Particularly under various flight operation conditions, the proposed algorithm performs well and its stability is sufficiently showcased. In a word, the proposed algorithm provides a candidate technique for predicting aircraft engine performance in dynamic behavior.