Time-Dependent Assignment Using a Neural Network

The assignment problem is a linear programming problem in which the goal is to determine the assignment of sources to destinations so as to minimize the cost. This arises in numerous areas, including assigning interceptors to threats in battle. Generally, the solution is very time consuming when the number of sources and destinations is large. Also, the problem is difficult when the cost of assignments is dynamically changing. Real-time solutions are needed, especially in a dynamic battle space like the boost-phase intercept of ICBMs using space-based weapons. Conventional solutions on conventional computer architectures are sequential and therefore not well matched to the problem. A dynamic parallel architecture, such as a neural network, can offer significant performance advantages and provide other benefits including increased solution accuracy in the presence of uncertainty of cost information and solution stability as the problem scope changes in real-time. The objective of this research is to design a neural network prototype for solving the time-dependent assignment problem. The performance of this network will be compared to conventional algorithms in solving a set of realistic space-based interceptor assignment problems.