Nonlinear neural network internal model control with fuzzy adjustable parameter

A novel nonlinear internal model control (NIMC) strategy based on neural network is proposed. The neural network is trained from system input-output data using a conjugate gradient algorithm to speed-up the convergence. The NIMC controller consists of a model inverse controller and a robust filter with single adjustable parameter. Two alternative control schemes, direct and indirect control, are discussed and improved. In the direct control, the neural network (controller) is used as a inverse process dynamics whose output is explicitly calculated as control action. On the contrary, the indirect control calculates the control action by directly inverting the network describing the process dynamics and thus constructs a rigorous inverse process model. To accommodate general uncertainty descriptions and ensure offset-free performance, a fuzzy neural network is proposed here. Two highly nonlinear processes, an exothermic stirred tank reactor and pH neutralization, are simulated to demonstrate the effectiveness of the strategy proposed. Extensions for measured disturbances are also presented