An interpretable machine learning model for trajectory prediction based on nonlinear dynamics mechanism constraints: applications for HVs

It is challenging for hypersonic vehicles (HVs) with strong nonlinear dynamics characteristics to achieve high-precision trajectory prediction. The un-interpretability of current prediction models and the difficulty in on-orbit data acquisition, high transmission costs and low data integrity bring huge obstacles to the accuracy and reliability of online prediction results. An interpretable modeling method is proposed by the physical block modeling, attention mechanism and mechanism constraints in the training process. Moreover, the binary encoding and inertial module are introduced to further improve the prediction accuracy and efficiency. The interpretability evaluation index is designed to quantitatively evaluate the degree of coincidence between the interpretable prediction model and the mechanism formulas, which proves the credibility of prediction results. The results show that the interpretable model has a better effect on the incomplete training set in terms of accuracy and efficiency. With an 8% incomplete training set, the interpretable model reduces the mean absolute error by 62.9%. After introducing the inertial module, the mean absolute error and the root-mean-square error are reduced by 40.1% and 46.0%. The developed interpretable model not only ensures the prediction accuracy, but also reduces the dependence on the training data and provides a reliable method for high-precision trajectory prediction.