On the chaos analysis and prediction of aircraft accidents based on multi-timescales

Abstract

Abstract Aircraft accident is an outcome of complex nonlinear and multi-scale phenomena, integrated together in some coherent manner. Based on chaos theory, aircraft accidents series of 1/1/1964 to 31/03/2018 are used for this study at different timescales(HM-scale, ETD-scale, EFD-scale and ED-scale). Firstly, the delay times and the optimal embedding dimensions of aircraft accidents series at four timescales are calculated, the phase space reconstruction for accident series are explored. Secondly, Lyapunov spectrums of four series are determined. Finally, the CSVR prediction model based on Support Vector Machine is introduced. A comparison of results reveals that the Largest Lyapunov exponents of four series are all positive, aircraft accident series have chaotic characteristic. There are no clear variation of the time delay, embedding dimension and the largest Lyapunov exponent along with the increase of timescale. The experimental results of prediction model show that the prediction error descends firstly and then ascends along with the decrease of timescale. The EFD-CSVR method is more accurate than other CSVR and simple SVR based on nMAE, mRMSE and mMAPE criteria.