Building multivariable predictive control models by process simulation and data regression

Abstract

This paper presents a multivariable model-based predictive control (MMPC) approach. It aims at processes such as distillation that have many difficulties such as multiple disturbances, long process lag and dead time and complex heat and material balance calculations. To implement the control strategy and control scheme recommended in the paper, a manipulated variable feedforward control model, a process performance prediction model and a process optimization model must be created. Well-planned rigorous process simulations can effectively and efficiently generate the data essential for model development. Statistical data regression is a powerful tool in selecting the model's independent variables and in establishing the quantitative relationships between the dependent variables and the independent variables. Process dynamics missing in the steady state simulation can be built into the models through theoretical analysis and practical process tests. When actually implementing the control system, the model accuracy can be improved simply by adjusting some pre-designed tuning factors. This approach is proven based on several successful installations at Arco gas processing facilities.