A self-calibration extended Kalman filter(SEKF)method was presented. Recursive algorithms of the SEKF were established for three nonlinear dynamic models with unknown inputs

such as unknown systematic error

gust and fault. In many nonlinear engineering cases

such as navigation

signal process

fault diagnosis

the conventional extended Kalman filter (EKF) cannot eliminate the effect of the unknown inputs

and maybe always lead to greater filtering errors or even diverge. The proposed SEKF is applied to compensate and correct the unknown inputs

and improve filtering accuracy. Numerical simulation shows that mean and standard deviation of state estimate errors of SEKF decrease to 1/12 and 1/4 respected to the conventional EKF

respectively

and the filtering accuracy is effectively improved. The SEKF method is simple to calculate and easy to apply in engineering.