Abstract
During the underwater motion of an axisymmetric body, unsteady shedding of the ventilated cavity causes severe pressure fluctuations on its downstream surface, affecting the motion stability. A traditional control system relies on monitoring data for posture adjustment, but may fail due to the hysteresis effect. To address this, a multi-scale online method is proposed to predict the unsteady pressure caused by cavitation shedding. This method decomposes the unsteady pressure into two scales: large-scale pressure, predicted using a medium support vector regression (SVR) model, and small-scale fluctuating pressure, predicted via a multi-round online deployment (MROD) method. The MROD method employs an offline-trained double-layer long short-term memory network, iteratively invoked to intermittently incorporate real-time data for advanced predictions. The prediction accuracy and speed of this method are influenced by key hyperparameters, including the input sequence length, output sequence length, real-time interval between time steps, and time step interval between consecutive real-time data inputs. Results show that both MROD and SVR models exhibit high prediction accuracy and robust generalization ability for predicting the small-scale fluctuating pressure and large-scale pressure, respectively. The proposed method achieves weighted mean relative errors below 1% for both interpolation and extrapolation of unsteady pressure, demonstrating its effectiveness in predicting unsteady pressure for axisymmetric bodies under unknown operating conditions. This high-accuracy prediction ensures stable motion of the axisymmetric body in complex marine environments.
Funding Information
  • National Natural Science Foundation of China (U22B6010)
  • National Natural Science Foundation of China (12472246)
  • International Partnership Program of Chinese Academy of Sciences (025GJHZ2022118FN)

This publication has 44 references indexed in Scilit: