Integral reinforcement learning based pursuit–evasion game of hypersonic flight vehicle with nonaffine nonlinear model and input constraints

This paper addresses the integral reinforcement learning (IRL) based optimal solution of the Hypersonic Flight Vehicle (HFV)’s pursuit–evasion (PE) game. The nonlinear guidance model of HFV is a nonaffine nonlinear model, and the input is constrained. The PE game is formulated as a nonaffine dynamic model, and a nonquadratic function is adopted to deal with the input constraints. The PE game is transformed into coupled Hamilton–Jacobi–Isaacs (HJI) equations, and then the IRL algorithm is adopted to solve the HJI. A novel pre-compensator method is proposed to transform the constrained PE game from nonaffine nonlinear systems into an unconstrained affine form, and then neural networks are utilized to obtain the optimal solution of the constructed unconstrained affine system. The existence of Nash equilibrium and the global asymptotic stability of the system are all discussed in relation to optimal control and differential games. Finally, this technique is applied to a PE game between HFV, and simulation results show that this method is very effective.