To address the difficulties and poor universality of traditional algorithms for accurate modeling of high maneuvering targets， this paper proposes an improved unscented Kalman filter （UKF） universal target tracking algorithm based on gated recurrent unit （GRU） prediction. The self-attention mechanism is introduced into the traditional GRU model， so that the model can better mine the kinematic equation of the target from the historical observation data and solve the problem that it is difficult to accurately model the maneuvering target manually. The learned model is applied to the UKF algorithm to achieve target tracking. Simulation experiments show that compared with traditional GRU models and classical algorithms such as UKF and interacting multiple model （IMM）， the proposed algorithm has better tracking accuracy and adaptability.