A neural network model based on differential-algebraic equations for nonlinear programming

A neural network model based on differential-algebraic equations for nonlinear programming is proposed. The penalty function method or barrier function method is used to convert a constrained optimization problem into a single unconstrained optimization problem by placing the constraints into the objective function. The resulting nonsmooth unconstrained penalty problem or barrier problem for finding an optimal penalty or barrier parameters is solved by a differential inclusion. A method for selecting a single or valued vector-field is presented. The global and local convergence properties of the new neural network model for nonlinear programming are analyzed. Examples are used to demonstrate that the network is both fast and more accurate than that of previous neural network models and classical methods