ESN-based combination method for aero-engine condition prediction

As a nonlinear time series prediction method, echo state network (ESN) attracts more attention because of its good approximation capability for the nonlinear system. Aiming at the characteristics of nonlinear time series in the aero-engine's condition prediction analysis, such as noise and presenting chaos, a combination method based on ESN was proposed. Firstly, the noise contained in nonlinear time series was reduced by the wavelet analysis. Then the training sample data were yielded via phase space reconstruction of the time series. After reducing the dimension of the training sample data by principal component analysis, all the remaining principal data were sent into the ESN prediction model. An actual dynamic pressure time series of aircraft power was conducted. The experiments compare the proposed method with traditional ESN prediction model on prediction accuracy and time cost. The results show that the prediction accuracy rate of proposed method within 5 steps and one-step are totally raised 66.97%, and the prediction of the mean square error, normalized root mean square error and normalized absolute error are also improved simultaneously. The proposed method is an effective nonlinear time series prediction method in practice.