A Discrete-time Projection Neural Network for Solving Convex Quadratic Programming Problems with Hybrid Constraints

A new discrete-time neural network is proposed for solving convex quadratic programming problems with hybrid constraints. Based on the projection operator and convex optimization technologies, a single layer discrete-time neural network with monotonic descent dynamic step sizes is constructed. It is proved that the equilibrium points of the discrete-time neural network are globally exponentially convergent to the optimal solutions of the programming problem. Moreover, an algorithm is given based on the proposed neural network and the scheme of backtracking step-size adaptation. Finally, the proposed algorithm is applied to three types of quadratic programming problems and the gas oven identification via a support vector regression algorithm. The numerical experiments are performed to show the correctness and effectiveness of the results in this paper.