Performance Analysis of Model Predictive Control For Distillation Column

Abstract

Model predictive control is an advanced process control method. It is a popular technique in chemical plants and oil refineries. Model predictive controller depends on dynamic model of the process and predicts the future output and so that the present input is optimized to avoid the future error. An optimization problem is solved over a prediction horizon P by regulating M control moves .Dynamic matrix control is a popular MPC method and it relies on the state space model of the plant. In this work, first we represent the DMC as an LTI system. The effect of tuning parameter on both first order and second order system is observed by calculating transient parameters like settling time, rise time, peak over shoot. Then the close loop poles are calculated for a specific FOPDT by varying different tuning parameters using the DMC algorithm. From the observation, effect of tuning parameters like P, M, w, N are summarized and a design rule for the parameter adjustment of DMC is proposed. Next a brief study on distillation column is provided and a mathematical model is also discussed. The design rule and control strategy of distillation column are discussed. The control of a distillation column by PID controller is performed for different tuning methods. In order to get stable response decoupling technique is used. Two different techniques like inverted and simplified decoupling are performed and a comparison between them is given by calculating transient parameters. The control of a distillation column by the MPC is also performed. A comparison between two controllers (PID and MPC) is discussed. The features of MPC like constraint handling, disturbance rejection, set point tracking is observed. Here different distillation process is taken and its response after using an MPC controller is observed. MATLAB (matrix laboratory) provides a numerical environment and fourth generation programming language. It provides matrix manipulation, plotting of function, data and implementation of algorithms. It provides a different tool box and Simulink models for process control and design.Model predictive control tool box provides functions, Simulink block for analysing, designing and simulating model predictive
control. Here user can provide control and prediction horizon, weighting factor and model length. The toolbox can guide the user regarding tuning parameters and it also facilitates softening of constraints.