Due to the particularity of the application scenarios of maneuver concomitant anti-aircraft weapon systems， it is difficult to accurately establish a flexural deformation model that adapts to changing working conditions， and the accuracy of transfer alignment cannot be guaranteed. To solve this problem， a data-free intelligent technology is adopted， and the flexural deformation model in the transfer alignment mathematical model is replaced by a neural network. The connection weight coefficients of the neural network are expanded to part of the unknown variables of the transfer alignment model. The nonlinear Kalman filter is used to estimate all the variables of the model in real time， so as to obtain the misalignment angle between the main and the sub-inertial navigation. Based on the attitude information of the main inertial navigation system， the high-precision speed alignment during the movement is completed. Simulation experiments show that this method can estimate the misalignment angle in a short time and complete the high-precision initial attitude binding of the missile inertial navigation system.