Neuro-Estimator Based Generic Model Control Of A Non-Linear Cstr Having Multiplicity

The control of a non-linear jacketed Continuous Stirred Tank Reactor (CSTR) with steady-state multiplicity is challenging due to its unstable nature. Generally, CSTR is operated near/at unstable equilibrium nodes, which decides the optimal productivity of the process. In this paper, a neural-estimator based non-linear control structure is developed for a CSTR possessing multiplicity. A Neuro-estimator based on feed-forward neural network has been designed to estimate the reactor concentration, which is often an imprecisely known parameter of the CSTR. We integrate the Neuro-estimator with a generic model controller (GMC) to develop a Neuro-GMC structure which utilizes the concentration estimated by the Neuro-estimator. Both servo and regulatory studies are performed to assess the effectiveness of the Neuro-GMC in controlling the reactor. Two additional control schemes, namely an extended Internal Model Control (IMC) and a standard PI controller, are designed to compare performance of the designed Neuro-GMC. Simulation results highlight that even in the presence of process-model mismatch, the Neuro-GMC yields better tracking and disturbance rejection characteristics.