Multivariable process control method and apparatus

Abstract

On-line control of a process is achieved by feedforward computations of manipulated variable setpoints modified during each of frequent optimization cycles by feedback trim determined by actual versus predicted effects of setpoint changes on the process controlled variables. Process, economic, contractual and equipment parameter values are continuously polled and collected and used to compute the optimal setpoints for the manipulated variables before the feedback trim is applied. Drift factors are added to the feedback trim to provide on-line calibration of key measuring instruments. Drifts in process performance are updated on-line periodically through self-tuning routines computed as calibration factors for predictor and control equations based on rigorous process simulations and actual plant performance. The equations take the form of polynomials in which each term contains an adjustable coefficient and one or more variable process parameter. The tuning factors are automatically applied at chosen time intervals to minimize the size of the feedback corrections computed during every optimization cycle. Tuning changes are effected by multiplying all terms of the predictor polynomials by the same factor. Effects of changes in process equipment are also accounted for on-line through calibration routines wherein individual coefficients in the polynomials are modified to reflect the equipment change in the apparatus. Optimization cycles typically occur every thirty seconds, and the system is capable of handling wide fluctuations in operational and economic parameters.