A Modified Reinforcement Learning Scheme For Controlling The Mean Arterial Pressure

The original reinforcement learning scheme comprises two networks, one performs a controller and the other stands for an evaluator. Based on temporal difference predictive techniques, the evaluative network predicts an external reinforcement signal and estimates a more informative internal signal to adapt a set of parameters of the controller. This paper introduces a modified reinforcement learning scheme to simplify the original scheme. The mean theme of the proposed scheme is that it replaces the fuzzy neural network (FNN) evaluator by a simple evaluator that depends directly on the environment of the process be controlled. Based on a performance index, the proposed evaluator outputs a reinforcement signal in the range of [-1,1] using two methods, one is fuzzy and the other is a discrete uniform of reinforcement signals. Compared with the original reinforcement scheme, the computational demand of the proposed scheme is relatively slight. That makes the proposed reinforcement scheme suitable for controlling real time intensive processes. The mean features of the proposed scheme are reflected in our simulation results on controlling the mean arterial pressure system.