Finite-Time Recurrent Neural Network Models for Quadratic Program Subject to Time-Varying Linear-Equality Constraints

To solve the quadratic optimization program, a novel kind of finite-time recurrent neural network (FNN) is presented to improve performance of convergent rate and convergent precision. Compared to the traditional neural network models, FNN is of limited value activation function, which makes the convergent error to equilibrium point within finite time. The activation function for the FNN is required for limited energy, which has a wider application fields. In addition, the finite-time attraction of the matrix differential equation is analyzed, and the experimental results show the FNN models can ensure the convergent rate to zero in finite time. Finally, FNN models are applied to the quadratic optimization problems and redundant manipulator repeatable trajectory planning. Numerical simulation results verify the superiority of the FNN dynamical method.