An Interpretable Data-Driven Learning Approach for Nonlinear Aircraft Systems With Noisy Interference
Rui Cao, Kelin Lu, Yanbin Liu
Abstract
This article discusses a learning algorithm for nonlinear aircraft systems, which targets the weaknesses of data-driven algorithms, mainly poor generalization ability and limited interpretability. It handles these constraints by integrating physical information with data-driven techniques. Referred to as the physics-informed SINDY (PI-SINDY) framework in this article, it improves the standard SINDY algorithm to tackle strongly time-varying nonlinear flight systems. This method incorporates the physical information described by the aircraft's differential kinematic equations into the SINDY algorithm and can also deal with the effects of measurement noise, making it more robust and practical. The proposed method displays higher robustness and generalization ability in comparison with the original SINDY algorithm and the WSINDy method, as confirmed by simulation results. Finally, we use the nonlinear system model learned with the suggested method for tracking control to supplement its efficiency.