Transfer Alignment Filtering Compensating Algorithm Based on Federal Neural Network

Focusing on the influence of various disturbing sources under air environment on transfer alignment (TA), a compensating algorithm based on federal neural network was put forward. Firstly, standard Kalman filtering structure was improved by regarding disturbing errors as measurement input. Then, the neural network was designed to form federal structure with two subsystems, which were used to train measurement input estimating error, output layer weight error and hidden layer weight error. Further, the training algorithm of the federal neural network was deduced and its stability was proved, which ensured low computing load and strong feedback capability. The online disturbing errors were efficiently predicted and aided to modified Kalman filter for accurate estimation of misalignment. Results of simulation experiments validate that, without knowing any priori information, the proposed algorithm could timely adapt to TA environment, predict and rectify disturbing errors, achieve fast convergence and high accuracy. It is feasible for air-to-air missile to implement rapid and high accuracy TA under hard air environment with different apparatus.