Application of Proper Orthogonal Decomposition and Neural Networks to UAV Task Assignment

A new and unique strategy is presented which enables UAV teams to optimize the use of their combined resources to accomplish their mission given the need for real-time task allocation. The traveling salesman problem is used as a benchmark for this class of combinatorial optimization problems. The basic premise is that there exists a model which can transform the solution from exponential time to polynomial time. Monte Carlo simulations were used to prepare a large data base consisting of a set of geographically distributed targets and their corresponding optimal task assignment. An integrated learning approach based on the Karhunen–Loeve decomposition and Artificial Neural Networks is applied to obtain a predictive model. The various stages of learning are detailed to provide an insight into the model building process. Results for simple cases are very promising. The next step involves scaling the solution for higher order of complexity.