Nonlinear neural network internal model control with fuzzyadjustable parameter

摘要

A novel nonlinear internal model control (NIMC) strategy based onneural network is proposed. The neural network is trained from systeminput-output data using a conjugate gradient algorithm to speed-up theconvergence. The NIMC controller consists of a model inverse controllerand a robust filter with single adjustable parameter. Two alternativecontrol schemes, direct and indirect control, are discussed andimproved. In the direct control, the neural network (controller) is usedas a inverse process dynamics whose output is explicitly calculated ascontrol action. On the contrary, the indirect control calculates thecontrol action by directly inverting the network describing the processdynamics and thus constructs a rigorous inverse process model. Toaccommodate general uncertainty descriptions and ensure offset-freeperformance, a fuzzy neural network is proposed here. Two highlynonlinear processes, an exothermic stirred tank reactor and pHneutralization, are simulated to demonstrate the effectiveness of thestrategy proposed. Extensions for measured disturbances are alsopresented