Model predictive neural control with applications to a 6 DOF helicopter model

We present a method for optimal control of MIMO non-linear systems based on a combination of a neural network (NN) feedback controller and a state-dependent Riccati equation (SDRE) controller. Optimization of the NN is performed within a receding horizon model predictive control (MPC) framework, subject to dynamic and kinematic constraints. The SDRE controller augments the NN controller by providing an initial feasible solution and improving stability. The resulting technique is applied to a 6 degree of freedom (DOF) model of an autonomous helicopter.