Data-driven and knowledge-driven prediction methods for ventilated cavities based on Gaussian process

Accurate prediction of ventilated cavity length serves as a cornerstone in stabilizing the shape of ventilated cavities and improving the attitude stability of the vehicle. The objective of this study is to establish both data-driven and knowledge-driven methods for predicting ventilated cavity length, utilizing freestream velocity and ventilation rate based on sparse data collected from water tunnel experiments. The positive correlation between ventilation rate and cavity length is additionally incorporated into the modeling process as engineering knowledge to guide the model's behavior. The prediction results indicate that, by constructing a joint covariance function combining knowledge and data, the cavity length prediction model achieves a predictive accuracy of 90% using only 50 sets of water tunnel experimental data, ensuring conformity with the physical relationship between ventilation rate and cavity length. The performance metrics include an average root mean squared error of 25.96 mm, reduced by 26.66%, an average mean absolute error of 20.92 mm, reduced by 24.04%, and an average R2 value of 0.7024, increased by 14%. This study provides guidance for knowledge and data fusion modeling in the field of underwater ventilated vehicles.