Abstract: Multi-layer model predictive control has become the mainstream method in industrial process control. Based on this control structure, original steady-state optimization was embodied in two main situations according to different desired values obtained from the operator or upper process optimization. An optimization problem with a compound objective function was proposed to calculate the target for MPC, which could degenerate into linear or quadratic form or the combination of both due to diverse process requirements. In order to ensure that ultimate optimal target was feasible and critical variables were not saturated, adjustment measure was taken when it was infeasible. Aiming at ensuring the consistency of variables between optimization implementation and MPC, the optimal target was transformed into incremental form. Simulation results of the constrained CSTR system showed that the optimization implementation layer provided appropriate optimal target effectively for MPC towards various process requirements, which demonstrated feasibility of the proposed method.

Key words: optimization implementation, model predictive control, compound objective function, incremental variable, primary variable

摘要： 多层结构的预测控制已逐渐成为工业过程控制领域的主流控制方案。在此控制架构基础上，根据操作工或工艺优化所给定期望值的不同，将稳态优化问题具体化为两种基本情况，并对此提出基于复合目标函数的优化问题，可针对不同过程要求退化为线性、二次或二者兼有的优化问题形式。为保证最优目标的可行性并在一定程度上避免关键变量饱和，对不可行的期望值适当调整。将所得最优目标增量化处理后送入模型预测控制动态控制层，确保了上下层之间变量传递的一致性。包含约束的全混槽反应器系统仿真实例表明，流程的优化实现层可针对不同的过程要求有效给出最优目标以便动态控制，说明了该优化流程的可行性。

关键词: 优化实现, 预测控制, 复合目标函数, 增量变量, 原位变量