Time-constrained guidance law for a ground-attack missile with time-vary speed

This paper investigates the issue of impact-time-constrained guidance problem for a gliding missile and proposes a machine learning-based approach. A three-hidden-layer Deep Neural Network (DNN) is adopted with each layer included 100 neurons to realize the accurate prediction of the time-to-go of proportional navigation guidance (PNG), and an analytical impact time constrained guidance (AITCG) law is developed using the outputs of the DNN. Then a bias term is developed to nullify the difference between the predicted time-to-go and its desired value. The main benefit of this approach lies in its accurate time-to-go prediction with DNN for a highly nonlinear system. Hence the impact time will be corrected in quite an efficient manner. Simulation results demonstrate that the trained DNN accurately estimates the time-to-go and the AITCG law can meet the need of the impact time control. Numerous Monte-Carlo simulations are performed to support our findings.