An Adaptive Neural IMC Design of Nonlinear Dynamic Processes

In contrast to usually applied Neural Network (NN) based controllers where the structure and fixed parameters of NN were obtained off line, here we present a fully adaptive Internal Model-based Neural Control design aimed to unknown nonlinear industrial processes with stable dynamics. The internal model of the control plant is implemented by the NN provided with a Stochastic Gradient Descent (SGD) learning algorithm. To cope with a high variability of process gain at different operational points and possible high errors in estimation of the corresponding sensitivities of nonlinear process model we proposed one practical solution to eliminate offset in a steady state at constant system inputs. Some illustrations and performance testing of the proposed adaptive NN controller are given by using an example.