A Novel Neural-Fuzzy Guidance Law Design by Applying Different Neural Network Optimization Algorithms Alternatively for Each Step

摘要

In this research, a novel neural-fuzzy guidance law by applying different neural network optimization algorithms alternatively in each step is proposed, such as the Gradient Descent (GD), SCG (Scaled Conjugate Gradient), and Levenberg-Marquardt (LM) methods are applied to deal with those parameter variation effects as follows: target maneuverability, missile autopilot time constant, turning rate time constant and radome slope error effects. Comparing with the proportion navigation (PN) and fuzzy methods are also made; the miss distances obtained by the proposed method are lower, and the proposed acceleration commands are always without polarity changes or oscillation at the final stage.