Stig Strand

Coordinator MPC for maximizing plant throughput

Abstract In many cases, economic optimal operation is the same as maximum plant throughput, which is the same as maximum flow through the bottleneck(s). This insight may greatly simplify implementation. In this paper, we consider the case where the bottlenecks may move, with parallel flows that give rise to multiple bottlenecks and with crossover flows as extra degrees of freedom. With the assumption that the flow through the network is represented by a set of units with linear flow connections, the maximum throughput problem is then a linear programming (LP) problem. We propose to implement maximum throughput by using a coordinator model predictive controller (MPC). Use of MPC to solve the LP has the benefit of allowing for a coordinated dynamic implementation. The constraints for the coordinator MPC are the maximum flows through the individual units. These may change with time and a key idea is that they can be obtained with almost no extra effort using the models in the existing local MPCs. The coordinator MPC has been tested on a dynamic simulator for parts of the Karsto gas plant and performs well for the simulated challenges.

MPC in Statoil – Advantages with In-House Technology

Abstract Statoil has used an in-house developed MPC since 1997. Several applications are running in closed loop today, with quite a good performance and average service factors of approximately 99%. All application projects have used internal resources only, which means that the competence has been aggregated within the company and that the best practice has been integrated in the MPC software. The total cost of software development is less than or similar to the alternative external vendor license fees, and the application project costs are definitely low.

INCREASED OIL PRODUCTION BY ADVANCED CONTROL OF RECEIVING FACILITIES

Abstract Slug flow from wells or multiphase flow lines from subsea satellite fields is one major disturbance at offshore oil and gas processing facilities. Slugging is an undesirable flow pattern with liquid blockages that can occur in multiphase flow lines transporting both gas and liquid. The original contribution in this paper is the successful application of advanced control for the inlet facilities, where active slug control for two 11.7 km long multiphase flow lines is combined with model predictive control to handle slugs entering the inlet separators.

Real-time optimization of a Cat Cracker unit

Abstract This paper describes the real-time optimizer (RTO) for the Residual Catalytic Cracker plant at the Statoil Mongstad refinery. The RTO completes a project for control and optimization of this unit. The RTO models are basically regressed on data from a plant test sequence of 60 planned experiments, along with data from ordinary operation. The optimization problem has about 20 independent variables and 40 process model equations. Process models are continuously compared to actual measurements and biases are updated at process steady-state conditions. The predicted economics (objective function) is calculated under the restriction of entire plant mass balance satisfaction, while the constraint predictions are based on delta values from the current operating point. The products are priced in two or tree stages depending on their use, for instance naphtha used as blending component has a higher value than naphtha sold as a product solely. A commercially available sequential quadratic programming (SQP) routine solves the optimization problem. The optimization system is coded in FORTRAN and implemented in a plant computer along with model-predictive control (MPC) applications, both interacting with the real-time process database. Optimized independent RTO-variables are implemented on the plant as set-points for the MPC controlled variables or as desired values of the MPC manipulated variables. Based on preliminary estimates and optimization runs during commissioning the expected project pay back time is two months.

THROUGHPUT MAXIMIZATION BY IMPROVED BOTTLENECK CONTROL

Abstract In many cases, optimal operation for a plant is the same as maximum throughput. In this case a rigorous model for the plant is not necessary if we are able to identify the bottleneck. Optimal operation is the same as maximum throughput in the bottleneck. If the bottleneck does not move, this can be realized with singleloop controller from the throughput manipulator to the bottleneck. However, if the bottleneck moves, single-loop control would require reassignment of loops which is undesirable. A better approach is then to use a multivariable coordinator controller since input and output constrains are directly included in the problem formulation.

IMPLEMENTATION OF MPC ON A DEETHANIZER AT KÅRSTØ GAS PLANT

Abstract Model predictive control (MPC) is implemented on several distillation columns at the Karsto gas processing plant, Norway. The paper describes the procedure in the implementation of MPC at a deethanizer using the SEPTIC * MPC tool, including design, estimator development, model development and tuning. For the deethanizer, the variance in the product quality has been reduced with about 50%. The number of flaring episodes has also been reduced. An increase in impurities has not been challenged yet, so the average reflux flow and steam consumption to feed ratios are almost unaltered. * SEPTIC: Statoil Estimation and Prediction Tool for Identification and Control

Coordinator MPC with focus on maximizing throughput

In this paper we suggest a “coordinator MPC” to perform dynamic real-time optimization (DRTO) on a plant. We consider the case where the plant economic criteria can be simplified to maximize the throughput in the plant. A measure for the distance to the bottleneck is formulated for a distillation column and the responses from feed to remaining capacity are expressed by experimental step-response models. The coordinator is demonstrated on a dynamic simulator and performs well for the simulated challenges.

Industrial Implementation of a Coordinator MPC for Maximizing Throughput at a Large-Scale Gas Plant

This thesis discusses plantwide control configuration with focus on maximizing throughput. The most important plantwide control issue is to maintain the mass balances in the plant. The inventory control system must be consistent, which means that the mass balances are satisfied. Self-consistency is usually required, meaning that the steady-state balances are maintained with the local inventory loops only. We propose the self-consistency rule to evaluate consistency of an inventory control system. In many cases, economic optimal operation is the same as maximum plant throughput, which corresponds to maximum flow through the bottleneck(s). This insight may greatly simplify implementation of optimal operation, without the need for dynamic optimization based on a detailed model of the entire plant. Throughput maximization requires tight bottleneck control. In the simplest case when the bottleneck is fixed to one unit, maximum throughput can be realized with single-loop control. The throughput manipulator should then be located at the bottleneck unit. This gives a short effective delay in the control loop. Effective delay determines the necessary back off from constraints to ensure feasible operation. Back off implies a reduction in throughput and an unrecoverable economic loss and should therefore be minimized. We obtain a rough estimate of the necessary back off based on controllability analysis. In some cases it is not desirable to locate the throughput manipulator at the bottleneck. To reduce the effective time delay in the control loop from the throughput manipulator to the bottleneck unit, dynamic degrees of freedom, like most inventories, can be used to reduce the effective time delay. In larger plants there may be several independent feeds, crossovers and splits that should all be utilized to obtain maximum throughput. The proposed coordinator MPC both identifies the bottlenecks and implements the optimal policy. A key idea in the coordinator MPC is to decompose the plantwide control problem by estimating the remaining capacity for each unit using models and constraint in the local MPC applications. The coordinator MPC is demonstrated by dynamic simulation and by implementation on a large-scale gas processing plant.

Model Predictive Control (MPC) of a Gas-Oil-Water Separator Train

This paper presents a simulation study of a state-space based Model Predictive Controller (MPC) applied to control the gas-oil-water separator train on Statoil’s Statfjord A crude oil production platform. The simulator is a fairly detailed one, used for several engineering purposes. The MPC-performance is compared to those of the existing basic control system and an LQG-controller. It is superior to both from different reasons.