A novel intelligent neural guidance law design by using adjoint method

Abstract

In this study, we propose a novel intelligent neural guidance law by applying several neural network optimization algorithms alternatively in each step, such as Gradient Descent (GD), Levenberg-Marquardt (LM) and Scaled Conjugate Gradient (SCG) methods. The missile turning rate time constant, radome slope error, initial heading error and the noise effects in the guidance loop (such as target maneuver, glint, and fading noises) are taken into consideration by using the adjoint simulation technique. Comparisons with the traditional proportional navigation (PN) method and those applying only one optimization algorithm for the cases of lower and higher altitudes are also made; note that the miss distances obtained by the proposed neural guidance law are always lower.