Neural networks and optimization problems

Artificial neural networks find wide applications in control engineering, image recognition, classification and optimization of different processes. However, for the solution of high order optimization tasks, efficiency of artificial neural networks may decrease because of high growth of expenses for the development of a composite artificial neural network and, especially, for its tuning for applied tasks. This paper presents the method of decomposition of high order optimization tasks into the series of lower order tasks. This allows one to significantly speed up the solution of the tasks with application of uniform artificial neural networks of parallel action. Decomposition procedure is formed using incidence matrices and vectors in conjunction with the threshold level of intensity of interconnections in the optimization system.