A new algorithm for the continuous trackingof the optimal economic operating conditions of an integrated chemical plant has been developed. The method does not make use of fundamental models of the plant but is based on an on-line search using experimental moves of the independent variables. The results of these moves provide the data for the identification of a dynamic process model. This model allows one to determine the gradient of the objective function and thus guides the next move of the independent variable eventually leading to the optimum. The advantages of the new technique over other methods are demonstrated in simulation examples:1) The speed of tracking is equally fast or faster than the relaxation speed of the system after a step change.2) The noise insensitivity allows safe tracking of the optimum despite significant measurement errors.As a key feature a decentralized form of the algorithm is developed making it suitable for distributed microprocessor implementation.
SCOPEDue to modelling inaccuracies and the influence of unmeasurable process disturbances the economically optimum operating conditions of a process can generally not be predicted but are determined during the operation by experimentation. This procedure encounters difficulties of two kinds: 1) Long experimentation times do not allow one to track a drifting optimum; and 2) Measurement noise hides the interdependencies of input and output variables and makes the reliable determination of the optimal search direction difficult. Starting from the early work of Draper and Li (1951) over Evolutionary Operation (Box and Draper, 1969) to the application of modern optimization techniques, the large number of attempts toward a solution which are reviewed below, gives an indication of the importance of the problem. The new approach suggested here is based on the identification of a dynamic system model and uses measurements during the transient and not only at steady state. This makes the algorithm faster and less noise sensitive. Simulation studies clearly indicate the superiority over previous methods. While only the open loop unconstrained search is treated in this paper, the second part of this series will elucidate the modifications necessary when constraints are encountered and how the optimizing controller can be interfaced with a regulatory controller.
CONCLUSIONS AND SIGNIFICANCEA new decentralized form of optimizing controller for integrated processing systems was developed and test results obtained from the simulation of a two CSTR example demonstrated its superiority with respect to speed and noise insensitivity. Both the speed of convergence and the degree of noise suppression can be influenced by several adjustable parameters of the optimizing controller. In general it was found exactly as for the tuning of filters that superior noise suppression slows down the parameter estimation and thus the speed of movement of the optimizing controller has to be reduced to avoid divergence. Good conservative a priori estimates of the controlle...
