Model predictive control having application to distillation

A method of controlling a system having manipulated variables to control process parameters of the system in response to deviations of the process parameters from target values related thereto:

• repeatedly executing a model predictive controller at a controller frequency;
• the model predictive controller having a dataset, containing records over a previous time period equal to a prediction horizon of states of the controls as manipulated variables and corresponding actual values of the process parameters as controlled variables and predicted values for the controlled variables, the model predictive controller also being programmed with step response models relating the manipulated variables controlled variables;
• during each execution of the model predictive controller, updating the dataset with actual current values of the manipulated variables, utilizing the dataset and the step response models to calculate prediction errors, applying the prediction errors to predictions as an off-set and calculating open and closed loop predictions over the prediction horizon and obtaining a set of move plans for movements of the manipulated variables to minimize the difference between the controlled variables and target values related to the controlled variables, and generating signals referable to initial movements contained in the set of move plans;
• the controlled variables including first and second controlled variables, the manipulated variables including a first manipulated variable having an effect on the first and second controlled variables and a second manipulated variable also having an effect on the first and second controlled variables and the step response models including first and second step response models relating the first manipulated variable to the first and second controlled variables, respectively, and third and forth step response models relating the second manipulated variable to the first and second controlled variables, respectively;
• the open and closed loop predictions being calculated for the first controlled variable and the first manipulated variable through superposition of the first step response model and the third step response model and with the second manipulated variable being used in connection with the third step response model as a first feed forward variable and the open and closed loop predictions being calculated for the second controlled variable through superposition of the forth step response model and the second step response model and with the first manipulated variable being used in connection with the second step response model as a second feed forward variable;
• and
• transmitting the signals to controllers used in setting the first and second of the controls to implement the initial movements of the first and second of the controls.