1995 Volume 31 Issue 7 Pages 908-915
For a control problem of a complicated nonlinear system with unknown dynamics, system identification method is a useful method, because the method makes clear the properties of the system. But this method with usual linear approximation model is not useful enough for the properties of natural phenomena, for instance atmospheric phenomena, earthquakes, etc., and of biological behaviors, for instance brain action, an organic action, etc. When such a complicated system is analyzed, it is not able to make precisely clear the properties of the system by analyzing properties of individual elements separated from the whole system, because such elements of such a complicated system have polysemy. In other words, in such a system, complicated interaction among the elements exerts an influence upon the system behavior.
For this problem, in this paper, a new model of complicated nonlinear systems with unknown dynamics is proposed, and an adaptation algorithm using the new model is presented. The model consists of a holon network in which the reciprocal action among the elements (holon) exists, and which is able to realize evolution of its performance adapting to its environment by changing network parameters. In this adaptation algorithm, the network structure is reconstructed by autonomous decentralized adaptation of the parameters keeping balance of two properties of holon, which are properties as the whole to the parts and as the parts to the whole. It is shown for some examples that this algorithm leads to a network structure adapting to the environment and that this model has sufficient generalization ability.
