Real-time nonlinear optimal control using neural networks

In this paper, a neural network based controller which optimizes a finite horizon quadratic cost function is developed for a class of nonlinear systems. The controller converges to its optimal value real-time eliminating the need for a priori knowledge of the nonlinearity and the initial conditions. The method makes use of the optimality conditions obtained from the Hamiltonian directly. These conditions are realized by a series of neural networks which converge to the optimal control iteratively in real-time. A nonlinear system to demonstrate its applicability is also included.