Kernel Adaptive Dynamic Filtering for Hypersonic Maneuvering Target Tracking

Without dependency on a prior model, kernel method provides a novel way of state estimation for nonlinear dynamic system with uncertainty. To perform maneuver target tracking without system model, a sliding-windowed adaptive kernel dynamic filtering method is proposed in this paper. The method performs Kalman filtering on a Reproducible Kernel Hilbert Space (RKHS) by estimating conditional embedding operator (CEO) as system state transfer function with training data. For system stochastic uncertainty, maximum correntropy criterion (MCC) is introduced for realizing kernel parameter optimization and the balance of estimation performance. The novel design of a sliding window is necessary for updating the estimates of the state transfer function online and to generate adaptability to unknown system dynamics. Such construction of the kernel Kalman filtering (KKF) is helpful to extend its application range from time-series signal processing to state estimation of uncertain dynamic systems. A typical scenario of hypersonic maneuvering target tracking is used for simulation. The results have demonstrated that the proposed method can realize effective model-free tracking to a target with nonlinear dynamical motion, showing the adaptability to non-cooperative target maneuver and better precision than typical model-based algorithms.