Single variable on-line adaptive optimisation of a water gas shift reactor

Abstract

A laboratory scale water gas shift reactor was optimised continuously with respect to each of two process variables using an on-line analogue computer. A function representing the instantaneous performance was defined in terms of both an inlet and an outlet measurement and its gradient was found by sinusoidal perturbation. The gradient estimate was used to drive the controller set point to the optimum position at which the gradient is zero. The influence of various controller parameters on the time-averaged value of the performance function was studied. For the temperature optimisation it was found possible to optimise satisfactorily with a perturbation period equal to a half of the system time constant and with optimum settings of the parameters the optimum was reached in about six time constants. The gradient estimate in the case of the steam flow optimisation was biassed by a frequency dependent term and this together with the particular shape of the objective function resulted in less efficient optimisation. An off-line mathematical model was formulated and used to predict the performance of the system and to extend the investigation of the controller settings. In the simulation study the best performance indices expressed as loss due to not being permanently at the optimum were 1·26 per cent for the temperature and 31 per cent for the steam flow optimisation. In each case the performance was averaged over a period of approximately fifty time constants.