| description abstract | Abstract. Accurate onboard modeling is fundamental for the development and implementation of advanced control algorithms in aero-engines. The onboard model based on dynamic coefficient method has been applied in engineering due to its good real-time performance. However, conventional dynamic coefficient methods fail to fully account for the uncertainties inherent in the actual flight process, thereby limiting modeling accuracy. To address this issue, this paper proposes an improved dynamic coefficient modeling method for turboshaft engines based on uncertainty perception. The proposed approach first employs wavelet filtering to preprocess flight data and introduces a state-perception-based steady-state data selection strategy to avoid incorrect or missing selection of steady-state data. Subsequently, a Gaussian mixture model (GMM) clustering technique is utilized to quantify and perceive data uncertainty by calculating data dispersion, thus enabling uncertainty-driven data grouping and modeling. On this basis, optimization algorithms are incorporated to further refine the dynamic coefficients and enhance the accuracy of the dynamic model. Simulation results demonstrate that, while maintaining low algorithmic complexity and data storage requirements, the proposed method improves the average steady-state modeling accuracy for gas generator speed, compressor outlet pressure, and turbine outlet temperature by 38%, 56%, and 73%, respectively, compared with conventional methods. The average dynamic modeling accuracy is also improved by 41.4%, 39.4%, and 50.6%. These results verify that incorporating uncertainty perception can effectively enhance the modeling accuracy and practical value of onboard models. | |
