Abstarct: This involves enlarging the size of the optimization problems that exist in practice. The necessary conditions of efficiency in the use of techniques that enable high-speed, very large problems solved with acceptable quality can be felt more than. Recently methods of optimization baxsed on artificial intelligence approaches have been developed remarkable success in solving optimization problems efficiently acquired. Methods such as Genetic Algorithms, Tabu Search, refrigeration simulation and neural networks, their ability to solve large problems have good action. Special rates available on the possible application of neural networks in a wide range of research has provided. It points to the possibility of learning and performance improvement baxsed on the input data point. It also allows parallel computations in a neural network is another advantage of the parallel hardware, enabling very large problems by this approach possible. In this thesis, we tried two different models of recursive neural network is presented to solve optimization problems in the traces. Analysis of uniqueness, stability and convergence of global solutions are examined and the performance of the proposed methods using several examples of second-order cone constrained variational inequality problems is shown. Finally, we provide conclusions and recommendations for future work.