A Modified Neural Network for Solving General Singular Convex Optimization with Bounded Variables

Singular nonlinear optimization problem has been the difficulty for optimization, which is frequently encountered in practical applications. People have been using numerical iterationmethods to deal with the singular problem previously, but the numerical instability and large calculation amount have not been able to be resolved. Based on neural network, this paper puts forward a continuity solution with any rank defect, by using the Augmented Lagrangian method and Projection to form a stable model. By using LaSalle's invariance principle, it is shown that the solution of the proposed network model is convergent. The numerical simulation also further confirmed the effectiveness of the method.