A Double Layer Neural Network for Solving Mixed-Integer Quadratic Optimization Problems

摘要

The main aim of this paper is to propose a double layer neural network (DLNN) for solving a class of mixed-integer quadratic optimization problems. It is proven that the DLNN is globally stable in the sense of Lyapunov and the output of the proposed neural network converged globally to an optimal solution. Compared with conventional Boltzmann Machine (BM) neural networks, the proposed neural network has a very small model size owing to its double layer structure. Furthermore, an application to power system showed that DLNN is effective. Numerical results shows that the DLNN is reduced by up to one-fifth compared to BM neural networks.