带有网络拓扑优化的分布式预测控制方法
Distributed model predictive control with optimal network topology for large scale systems

通常在大系统中, 全局信息优化的系统, 其性能要高于局部信息优化系统. 全局信息优化的算法由于大系统的复杂程度往往不可行. 所以通常会用分布式算法来解决此类问题. 在分布式算法中, 为了获得更好的系统性能, 要尽可能多的采用更多的信息信息交换, 然而这样会带来信息网络的负担增大. 本文在预测控制性能指标中引入通信代价, 并提出了一种随着系统状态变化的通信网络拓扑切换方法. 文中给出了该算法在供水管网动态模型中的仿真结果, 表明本方法的可行性.
Generally, the performance of a large-scale system in global optimization is better than that in local op- timization. To obtain a better performance, we need more information interchanges. However, in a distributed model prediction algorithm, this will increase the workload for an information network, making global optimization inapplicable in a large-scale system with complexity. To deal with this problem, we introduce the communication cost in the perfor- mance index and develop a novel method for switching the communication-network topology according to variations in system states. This method has been applied to the simulation experiment of a dynamical model of a water supply network system, demonstrating the feasibility of the proposed algorithm.