PELT-Phase LSTM Method for Mach Number Prediction of Wind Tunnel Flow Field

Abstract

 The stability of Mach number plays a key role in wind tunnel flow field performance. In order to facilitate the direct processing of continuous transonic wind tunnel 3D data and improve the accuracy of Mach number prediction, this paper proposes a new method of Mach number prediction based on PELT-Phase LSTM model. Firstly, the data are divided into intervals using the PELT change-point detection algorithm, and then divided into three phases according to the Mach number trend of each interval by batch, and finally predicted by LSTM model in phases. The results demonstrate that, compared to the global model, the proposed model exhibits improved prediction accuracy at both Mach number 0.3 and Mach number 0.8, with corresponding improvements in RMSE of 6.703% and 33.3995% respectively, which proves that the model has very high prediction accuracy and good stability performance.