Dag Ljungquist

State—space predictive control

Abstract This paper deals with a predictive control strategy based on state—space models. Important issues concerning inherent model identification and optimal control computation are briefly discussed. Predictive control relies heavily on a model with satisfactory predictive capabilities. An off-line identification procedure must be accomplished to obtain a proper model structure and a parameter set, which is required for on-line adjustment. The control calculation is based on a general performance index and parameterization of the control variables in a nonlinear model, which includes the relevant constraints. This results in a finite-dimensional optimization problem which can be repetitively solved on-line. Simulation studies on two very different, typical industrial processes are presented. The simulations show that this MPC technique offers a major improvement in the control of many industrial processes.

Predictive Control based upon State Space Models

Repetitive online computation of the control vector by solving the optimal control problem of a non-linear multivariable process with arbitrary performance indices is investigated. Two different methods are considered in the search for an optimal, parameterized control vector: Pontryagins Maximum Principle and optimization by using the performance index and its gradient directly. Unfortunately, solving this optimization problem has turned out to be a rather time-consuming task which has resulted in a time delay that cannot be accepted when the actual process is exposed to rapidly-varying disturbances. However, an instantaneous feedback strategy operating in parallel with the original control algorithm was found to be able to cope with this problem.

Catalytic cracking models developed for predictive control purposes

The paper deals with state-space modeling issues in the context of model-predictive control, with application to catalytic cracking. Emphasize is placed on model establishment, verification and online adjustment. Both the Fluid Catalytic Cracking (FCC) and the Residual Catalytic Cracking (RCC) units are discussed. Catalytic cracking units involve complex interactive processes which are difficult to operate and control in an economically optimal way. The strong nonlinearities of the FCC process mean that the control calculation should be based on a nonlinear model with the relevant constraints included. However, the model can be simple compared to the complexity of the catalytic cracking plant. Model validity is ensured by a robust online model adjustment strategy. Model-predictive control schemes based on linear convolution models have been successfully applied to the supervisory dynamic control of catalytic cracking units, and the control can be further improved by the SSPC scheme.

State space model predictive control of a multistage electrometalurgical process

Abstract The paper deals with the proposed application of a novel technique for Model Predictive Control to a multistage electrometalurgical process. The novel control technique is based upon high speed repetitive simulation of a nonlinear state space model of the process including relevant constraints, and searching in a parameterized control space by an efficient optimization routine until an optimal set of control actions has been found. This MPC–technique is not limited to linear processes with quadratic objective functionals and is therefore believed to offer major improvement to the control of many industrial processes where the standard, linear control solutions fail because of nonlinearities and constraints in the process system.

Catalytic Cracking Models Developed for Predictive Control Purposes

Abstract The paper deals with state-space modeling issues in the context of model-predictive control with application to catalytic cracking. Emphasize is placed on model establishment, verification and online adjustment. Both the Fluid Catalytic Cracking (FCC) and the Residual Catalytic Cracking (RCC) units are discussed. Catalytic cracking units involve complex interactive processes which are difficult to operate and control in an economically optimal way. The strong nonlinearities of the FCC process mean that the control calculation should be based on a nonlinear model with the relevant constraints included. However, the model can be simple compared to the complexity of the catalytic cracking plant. Model validity is ensured by a robust online model adjustment strategy. Model-predictive control schemes based on linear convolution models have been successfully applied to the supervisory dynamic control of catalytic cracking units, and the control can be further improved by the SSPC scheme.