As the nonlinearity and complexity of a nonlinear system increase, it may be more difficult to construct a controller by the mathematical control theory. In such cases, it is very effective to construct the controller by using Neural Networks (NNs), because NNs have the capabilities of coping with the nonlinearity and complexity of the nonlinear systems. NN controllers are constructed through learning to minimize a criterion function under certain environments of the system. But NN contollers may not work well under a very different environment from the environment at learning stage. In other words, for example, NN controllers are usually made without considering the changes of the environments because NN controllers do not have a means to suppress or to activate their influences. So, in the case that the environment changes, NN controllers do not work well. In this paper a suppressed or activated control system design methods for the changes of the environments of the system are discussed using second order derivatives of Universal Learning Network.
抄録全体を表示