Supervised learning control of a nonlinear polymerisation reactor using the CMAC neural network for knowledge storage

The CMAC neural network is an adaptive system by which complex nonlinear functions can be represented by referring to a lookup table. In this paper, this network is applied to the state estimation and learning control of the continuous-stirred tank reactor (CSTR), which is a widely used polymerisation reactor system. The study involves the estimation of the online unmeasurable state and the realtime setpoint tracking of the two-input/two-output CSTR system. Simulation results show that the CMAC-based method is strong in self-learning and easy to realise, and is helpful for improving the nonlinear control performance.