The Algorithm And Strategy For Two-Layered Model Predictive Control In Large-Scale Industrial System

Two-layer model predictive control (MPC) has solved the problem about whether set-points are chosen appropriately, and has set the optimization problem which concerns system’s stability, safety and economic efficiency and set-point tracking problem apart, which makes a clearer hierarchy of optimization and control. However, two-layer model predictive control is more usually used in medium or small scale industry system other than large scale system because of the disadvantage of its large amount of on-line calculation. The theoretical research about two-layer model predictive control’s optimization and control is inadequate, thus its application in industry process lacks appropriate theoretical guidance.The work in this paper is application-oriented, we’ve found out that control inputs’ number and prediction horizon’s length are key reasons why two-layer mpc’s calculation is so huge, and we’ve come up with a conclusion that its computational complexity is proportional to cube of control inputs’number and prediction horizon’s length’s product. We’ve presented a new control strategy named centralized optimization and decentralized control for two-layered predictive control through modification of the very algorithm’s optimization and control structure based on the conclusion mentioned above aiming at decreasing the huge amount of calculation in large scale systems and maintaining its global optimization performance. And we’ve also analyzed the strategy’s economic performance and robustness when exist model mismatch and unmeasured disturbance.In this paper, the main work and achievements are as follows:1. In view of the flaw that artificially set expectations for traditional predictive control, elaborated to the two-layered MPC algorithm of linear programming including the steady-state target calculation automatic optimization method combined with dynamic control tracking the set value, and the control algorithm is described on the perspective of structure, leading to the inner relationship between optimization and control for research. 2. Two-layered MPC has high on-line complexity and long on-line solving time on large-scale systems. Through theoretical analysis of computational complexity and implemen-tation of simulation for the steady-state target calculation layer and the dynamic matrix control layer, we have come up with a conclusion in this paper that number of control inputs and control horizon are the factors that affect algorithm’s computational efficiency most, and the desired algorithm’s time complexity is proportional to the cube of product of control inputs’ number and control horizon. And it is valuable to controller’s design in industry process.3. To reduce the on-line computational complexity of MPC for large-scale systems and maintain the global optimization performance, a new strategy of two-layered MPC was proposed. In order to keep the global optimization and diminish the on-line calculation of dynamic control, the strategy optimizes overall process in steady-state target calculation and realizes decentralized control by dividing the dynamic control into multiple subsystems. Moreover, to decrease the influence of intermediate variables and improve control performance of the system, feed-forward controllers are added between the subsystems for disturbance compensation.4. To decrese the influence of model mismatch and unmeasured disturbance.in the two-layered MPC, adds the steady state error to the steady-state target calculation economic performance target function, thereby eliminating the influence of the steady state prediction error, And through the model mismatch and unmeasured disturbance situation carries on the simulation analysis of the strategy performance.