Neural network for hierarchical optimization based on dynamic programming

A specific neural network is devised for solving hierarchical optimization problems of discrete time dynamic large scale systems in the paper. The whole neural network is itself of hierarchical structure that it is composed of coordination network in higher layer and local optimization networks in lower layer. Making use of the dynamic programming model which fits the formulation of the problem, the structure of the whole neural network can be made much more compact than other cases. Using a technique to imbed the dynamic equations of subsystems into the models of the corresponding local optimization networks, the dimension of the whole network is reduced greatly. It is proved that the dynamic behavior of the network is asmyptotically stable such that starting from any initial state, the state of the network evolves eventually to an equilibrium which corresponds to an optimal solution to the large scale optimization problem. Simulation shows high performance of the network. Copyright (C) 1998 IFAC.