分布式预测控制全局协调及稳定性分析

实际的化工过程系统维数都较高,对系统进行关联分解并使各子系统进行协调来实现整个大系统全局最优是必要的。对于关联作用存在反馈的化工过程大系统,分散式预测控制算法和基于串联过程推导的邻域优化分布式预测控制算法都不适用,因此在这两个算法的基础上推导出约束条件下基于全局协调的分布式预测控制算法。针对分解后得到的子系统,假设子系统间关联信息的传递存在一个采样时间的滞后,建立每个子系统的预测模型时考虑滞后的关联信息;建立子系统的目标函数时,综合考虑所有关联子系统的输入和输出对本子系统的关联作用;每个子系统滚动优化并行求解各自的最优控制作用。然后,在一定条件下分析了基于全局协调的分布式预测控制算法与集中预测控制算法的一致性,并说明了闭环系统的全局稳定性。最后,通过对Shell公司重油分馏塔和TE过程两个例子进行仿真并与其他算法进行比较,验证了本文提出算法的可行性和有效性。

Global coordination and stability analysis for distributed model predictive control system

With the high dimension in chemical processes, it is necessary to coordinate the sub-systems to achieve optimality of the global system after decomposing the large-scale system.The on-line optimization of the whole system is decomposed into several small sub-problems, significantly reducing the computational complexity in model predictive control of the large-scale system.Decentralized model predictive control and the neighborhood optimization algorithm for distributed model predictive control based on the cascade process are not suitable for the chemical process containing interactive feedback. Therefore, by taking the system constraints into account, the global coordination algorithm was deduced based on those two optimization algorithms.The large-scale system was decomposed into sub-systems with lower dimension.Assumably there was one sample time delay communication between interacted subsystems.For each sub-system, the predictive model was constructed and the overall objective function was formulated by considering the interactions of the sub-systems.The optimal control of each sub-system was achieved.Then, the consistency for distributed model predictive control based on global coordination and centralized model predictive control was analyzed under certain conditions.The global stability of the closed loop system with distributed model predictive control based on global coordination was also obtained. The proposed algorithm was tested by simulation study of Shell heavy oil fractionator and TE process. When compared with other algorithms, it was feasible and effective.