Remarks on neural network controller for a inverse dynamics of many-to-one plant

Neural networks have excellent characteristics such as a learning capability and a flexible structure. Several types of a neural network controller have been studied in order to incorporate these characteristics in servo controllers. These neural network controllers are expected to apply to any nonlinear plant. However, the input value of some nonlinear object plants is not one which corresponds to one output value. Most neural network controllers have to express the inverse dynamics of many-to-one plants. However the neural network output can only express one value. In this case the neural network appears to express only a part of the inverse dynamics. Therefore, one should investigate the characteristics of the neural network controller for a many-to-one plant. The author selects two types of neural network controller for investigation. One was proposed by K.S. Narendra and it has the identification stage and the control stage. The other controller is proposed by the author and it learns the inverse dynamics of the object plant in cooperation with control. Simulation results confirm the characteristics of these controllers when they are applied to a many-to-one plant