Control by a New Policy- and Experience-Driven Neural Network to Follow a Desired Trajectory

In previous paper, the concept of a policy- and experience-driven neural network (PENN) was proposed to provide adaptive and self-tuning features for process control. Rule-based global policies give general control information covering the whole control space, whereas local experiences achieved by previous runs give detailed information about the process features. There the setpoint was assumed to be constant. In the paper, the case where the setpoint is changing with time is discussed and a very simple network is proposed. The results for a level-control experiment by simulation show that this method has high potential for wide application