A One-Layer Dual Neural Network with a Unipolar Hard-Limiting Activation Function for Shortest-Path Routing

摘要

The shortest path problem is an archetypal combinatorial optimization problem arising in a variety of application settings. For real-time applications, parallel computational approaches such as neural computation are more desirable. This paper presents a new recurrent neural network with a simple structure for solving the shortest path problem (SPP). Compared with the existing neural networks for SPP, the proposed neural network has a lower model complexity; i.e., the number of neurons in the neural network is the same as the number of nodes in the problem. A simple lower bound on the gain parameter is derived to guarantee the finite-time global convergence of the proposed neural network. The performance and operating characteristics of the proposed neural network are demonstrated by means of simulation results.