A Neural Controller Using Imc Strategy

Abstract

In this paper a nonlinear Internal Model Control (IMC) strategy based on a modified NARMA model is proposed. The IMC controller consists of a model inverse controller and a robustness filter with a single tuning parameter. Under such a controller, the control quality and the robustness can be influenced in a direct manner. From the input-output observation data of the plant, a neural network is trained off-line using Back-Propagation algorithm to emulate the feed-forward dynamics of the plant. This neural plant model provides a simple check of the model invertability, which plays a critical role in the IMC strategy. Using the same network used in the feedforward model of the plant, an exact model inverse is obtained directly assuming invertability of the plant model. The exact inverse of the plant model is very important to achieve offset-free performance. Simulation examples demonstrate the simplicity of the design procedure and the good performance characteristics of the proposed nonlinear IMC controller. The examples show that the proposed controller is superior to conventional PID controller and it can also be applied to systems having time delay.