Research Of Model Predictive Control And Operation Optimization Method In Process Industry

Iron and steel industry, petrochemical and other industries belong to a class of chemical reaction-based process industry. The production processes of process industry generally have the following characteristics:difficulty of achieving optimal operation, difficult detection of disturbance parameters, complex production conditions, and interrelation of multiple manipulated variables, which make the control of these production processes extremely complex. Although the model predictive control (MPC) has been successfully applied in many production processes in process industry, it is still often faced with the above problems. Based on the background of process industry, we studied the problems of operation optimization, unmeasured disturbances, model mismatch and strong coupling for MPC.1) Aiming to eliminate the influences of model uncertainty on the steady-state target calculation in integrating process, we presented an operation optimization method based on "point" model and a method determining the existence of feasible solution for steady-state target calculation. The operation optimization method resolves the steady-state optimization problem of integrating processes under the framework of two-stage structure, which builds a steady-state predictive model based on "point" model for integrating process, and compensates the error between "point" model and real process in each sampling interval. Simulation results indicate that the steady-state predictive model can predict the future outputs of integrating variables accurately.2) For the effect of the unmeasurable disturbance in MPC for integrating process, an adaptive MPC method was proposed. Firstly, RELS (Recursive Extended Least Squares) algorithm is used to estimate the unmeasurable disturbance; secondy, the proportion of prediction error caused by the disturbance is calculated; lastly, twiddle factor of MPC is updated real-time to realize the adaptive predictive control. The effectiveness of the adaptive MPC is illustrated by its successful application in a VCM distillation process.3) Considering the condition that the steady-state prediction equation is difficult to establish for integrating process, we proposed a method based on integration of the steady-state optimization and dynamic optimization. This method establishes the steady-state prediction model that can reflect dynamic execution process of the manipulated variables; the input increment sequences of multi-step prediction are regarded as the decision variables, a quadratic programming model with inputs, outputs and input increment constraints was developed. Simulation examples demonstrated that the method can effectively solve the steady-state optimization problem for integrating process when economical optimization of the inputs and the outputs are considered.4) For the model mismatch problem existing in the long-running case for MPC, we presented a semi-automatic model predictive control (SAMPC) based on model reference adaptive identification (MRAI) algorithm. The method used a semi-automatic mode, which adds a persistent excitation signal to the controlled plant in the case of model mismatch. The transfer function model of the plant is identified by improved model adaptive identification algorithm. The simulation results show that the dynamic characteristics of identified transfer function model are much clearer and more convenient for analysis and modification. FSR model after inverse Laplasse transform is smoother and can eliminate the offset caused by model errors.5) For the condition that the controllability of the strongly coupled linear time-invariant (LTI) system is very poor, we presented a distributed model predictive control (DMPC) method to improve part of the dynamic performance of the system. The controllability structure of DMPC is designed based on a branch and bound algorithm. In addition, the topology of communication network of DMPC is designed according to the degree of correlation between the loops. The simulation shows that the settling time of distributed model predictive controller is significantly shorter than that of centralized model predictive controller, and control accuracy of distributed model predictive controller is also obviously improved than that of decentralized model predictive controller.