A new discrete-time neural network for quadratic programming with general linear constraints

Abstract

This paper presents a discrete-time neurodynamic model to solve linear and quadratic programming with respect to linear equality and inequality constraints. The new model is obtained by using an auxiliary variable, and can be seen as the generalization of a neural model for bound constraints in the literature in the sense that bound constraints limit a linear function of the desired variable. The proposed neural solution is proved to be stable in the sense of Lyapunov and converges globally to the optimal solution of the given minimization by proper adjustment of a parameter. The model is further simplified for the case that the equality constraints entails a full row-rank linear mapping. The proposed neural solution is comparable with the state-of-the-art in terms of both the number of operations in each iteration and the required components for its circuit implementation. The experiments confirm the reasonable performance of the proposed neuaral network.