Data-Driven Prediction Method of Tactical Missile Aerodynamic Characteristics

Using the engineering algorithm, the aerodynamic characteristics data set for tactical missiles with over 3000 shapes has been established. The adoption of four machine learning algorithms, including gradient boosting decision tree (GBDT), XGBoost, multi-task learning neural network (MTLNN) and embedded Physics Informed Neural Network (PINN), is made. A proposed method for predicting the aerodynamic characteristics of tactical missiles is based on data. Test set shows that the deep learning algorithm is better than the classic machine learning algorithm. The synthetic deviation of normal force and pitching moment coefficients obtained by the latter three machine learning algorithms is less than 5%; Due to the introduction of physical mechanism between aerodynamic characteristics parameters in PINN model, not only the prediction accuracy is improved, but also the dependence of the model on the number of training samples is reduced. The PINN model has strong generalization ability and high prediction accuracy can still be obtained for data outside the training samples. Among the three machine learning algorithms, PINN model has the best prediction effect, and the comprehensive deviation of normal force and pitching moment coefficient is 0.72% and 1.05%, respectively. A new idea for the aerodynamic design of tactical missiles is provided by this prediction method.