Nonlinear control based on an improved neural predictive control scheme

Generalized neural predictive control is a kind of receding horizontal control method, in which the minimum cost function is used to optimize the control input, but large calculation is needed in the minimization of the cost function. In this paper, time-delay neural network is imposed in the neural predictive control, and the new updating algorithm can return many derivative in Jacobian or Hessian at different time step, which improves the learning algorithm of the traditional neural predictive control in real-time control speed. The simulation results indicate the advantage of the proposed scheme in realtime control speed.