(598c) A Stable Direct Inverse Neural Network Control of An Isothermal Continuous Stirred Tank Reactor with Input Multiplicities


In this paper, a stable direct inverse Neural Network  controller (DINNC)   is designed for an Isothermal continuous stirred tank reactor (CSTR) with input multiplicities. The Continuous Stirred Tank Reactor exhibits input multiplicities in space velocity on concentration. i.e., two values of space velocity will give the same value of concentration. Processes with input multiplicities may experience unstable response, oscillatory and switch over to lower stable input with conventional PID controller. In this work, the direct inverse NN controller is designed separately at lower and higher inputs and the  DINNC controller  output which is nearer to the operating is selected for the implementation on process. The Performance of direct inverse neural network controller has been evaluated through simulations at lower and higher input steady states. The response of DINNC is found to be faster than conventional PI controller. Interestingly, the present DINNC controller gives stable response whereas the PI controller designed for lower input become unstable response at higher input steady states.  Thus, direct inverse neural network control is found to overcome the control problems i.e. instability due to the input multiplicities. The present controller (i.e., DINNC)  results in some offset due to highly nonlinearity of the CSTR system and the offset is eliminated by adding an integral action.