Authors: Umesh Mathur, Barry L. Burr, Robert D. Rounding, Daniel R. Webb, and Victor L. Rice
Abstract: Deploying adaptive multivariable controllers at two light hydrocarbon plants improved operations and profitability significantly, generating project paybacks within a few months. Traditionally, implementing model-predictive controllers (MPC) in the process industries has required creation of a fixed, linear, dynamic model that relates changes in each input to all outputs. Further, the vast majority of past projects described in the literature have been executed using extensive step tests to develop a linearized control model, using process model identification techniques. Such deliberate tests can be quite costly, disruptive, invasive, and lengthy in duration – often lasting many weeks or months in a large unit. We should mention that both of the plants described herein would have been be very difficult, if not impossible, to subject to extensive step tests in the traditional manner. This is because they suffer very large measured and unmeasured disturbances, have very long settling times, and because the disruption caused by step tests would be quite unacceptable for reasons of safety and product quality.
We describe our experience in using economically optimal, plant-wide, adaptive multivariable controllers for stabilizing operations and improving profit without use of ANY step tests, and present two case histories: a lean oil gas absorption plant, and a refrigerated fractionation complex. These units obtain feed from pipelines and ship products directly to pipelines with no intermediate storage.
We discuss the manifold benefits observed from use of an adaptive control approach in both instances: (1) The adaptive nature of the controller enabled it to maintain excellent performance in the face of changing process dynamics, with very large and frequent feed rate disturbances exceeding 200% within 30 minutes in one of the units. In contrast, fixed model MPC controllers are generally incapable of handling such large disturbances. (2) The particular software package chosen in this effort required only that the user specify the steady-state gains for each manipulated and disturbance variable against each controlled variable. These were determined without use of traditional step tests. (3) The projects were executed with negligible impact on normal plant operations using analytical process simulation models, developed in an office environment, with reliance only on plant design / historical data to provide the needed MPC controller inputs. (4) The adaptive MPC controllers showed excellent stability and their performance was deemed to be very good, no modeling or other tuning changes being required for over a year, despite huge intervening changes in product values and utility costs in one of the cases. Operator acceptance has been exceptional.
In this work, we developed and calibrated steady-state simulation models against plant data for the processes in question and used these to develop the required MPC controller gains. The chosen MPC controller synthesizes the process dynamics on-line and also makes model adjustments automatically as conditions change. The resulting quality of control was extremely satisfactory compared to pre-project conditions. In both cases cited, a single MPC application was used to achieve economically optimal control of the entire plant, thereby enabling the controllers to adjust operations dynamically to alleviate all constraints.