Mitsuhiro Kaneda

A design of neural-net based predictive PID controllers

PID control schemes have been widely used for various real control systems. However, in practice, since it is difficult to find suitable PID gains, a lot of research has been reported with respect to tuning schemes of PID gains. In this paper, a design scheme of neural net-based PID controllers is proposed for nonlinear systems. The modeling error between a linear nominal model and a controlled object is compensated by a multilayered neural network whose outputs are given as additive PID gains. Furthermore, the fixed PID gains are calculated for the linear nominal model based on a generalized predictive control scheme.

A learning algorithm for a neural network in a linearlizer for nonlinear systems

The purpose of this study is to give a design method of a linearlizer by using a neural network (NN) with off-line learning algorithm. This linearlizer works so that the input-output property of the augmented system which consists of the system and the NN may be equivalently equal to that of the linear model. The learning of the NN is performed off-line by using the input and output data of the system. Finally, a numerical simulation is demonstrated to illustrate how to use it in the control problem.