协作式分布式模型预测控制算法通信策略研究

发布时间：2018-01-10 11:19

本文关键词：协作式分布式模型预测控制算法通信策略研究　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：在处理大规模系统的控制问题中,分布式控制平衡了集中式控制优良的控制性能和分散式控制的安全性,是一种较为理想的控制方式。模型预测控制(Model Predictive Control)凭借其对约束的处理能力和系统未来行为的预测能力,广泛应用到分布式控制中,形成了分布式模型预测控制(Distributed Model Predictive Control)。协作式分布式模型预测控制算法凭借全局的优化范围,可以实现帕累托(Pareto)最优解,但该优异控制性能的实现依赖于较高的通信负担,尤其当子系统的个数较多时,通信负担会十分沉重,阻碍协作式分布式模型预测控制在实际应用中的发展。针对协作式分布式模型预测控制算法,本文在减少其高通信负担方面开展了研究,主要研究内容包括:(1)提出了一种基于层次分解的分布式模型预测控制算法,根据系统的通信结构,采用解释模型法和回路分解算法,将子系统划分到若干个连通集内,并定义各连通集的层次顺序,最后,按照连通集顺序依次运算求解各子系统的最优输入。由于子系统只需要与连通集内的其他子系统通信,不需要和连通集外的子系统进行通信,此协同策略降低了通信负担;(2)提出了一种基于事件触发器的分布式模型预测控制算法。引入0-1变量来表示两个子系统间是否需要耦合,构造动态耦合模型,并根据该模型提出综合控制性能和通信代价的多目标优化命题,耦合结构的优化会改变原命题中耦合部分的结构,通过忽略一些子系统间的耦合作用,这些子系统之间不再需要进行通信,降低了通信负担。为了降低计算负担,该算法从系统的稳定性出发构建了一个事件触发器,当触发器触发时,求解一个包含耦合结构决策和控制性能的优化命题,更新一次耦合结构,否则,保持上此优化后的耦合结构不变,只求解控制性能的优化命题,计算最优的控制作用。相比于基于动态耦合模型的分布式模型预测控制算法,该算法降低了计算复杂度。
[Abstract]:In dealing with the control problem of large-scale systems, distributed control balances the excellent control performance of centralized control and the security of decentralized control. Model Predictive Control (MPC) is an ideal control method. Model Predictive Control (MPC) relies on its ability to deal with constraints and predict the future behavior of the system. It is widely used in distributed control. Distributed Model Predictive Control is formed by distributed Model Predictive Control. . collaborative distributed model predictive control algorithm based on global optimization. The Pareto optimal solution can be realized, but the implementation of the excellent control performance depends on the higher communication burden, especially when the number of subsystems is more, the communication burden will be very heavy. The development of collaborative distributed model predictive control in practical applications is hindered. Aiming at collaborative distributed model predictive control algorithm, this paper studies how to reduce the high communication burden of collaborative distributed model predictive control algorithm. The main research contents include: (1) A distributed model predictive control algorithm based on hierarchical decomposition is proposed. According to the communication structure of the system, the interpretive model method and the loop decomposition algorithm are adopted. The subsystem is divided into several connected sets and the hierarchical order of each connected set is defined. Finally. The optimal input of each subsystem is solved according to the order of the connected set. Because the subsystem only needs to communicate with other subsystems in the connected set, there is no need to communicate with the subsystem outside the connected set. This cooperative strategy reduces the communication burden; (2) A distributed model predictive control algorithm based on event trigger is proposed. The 0-1 variable is introduced to indicate whether the two subsystems need to be coupled or not, and the dynamic coupling model is constructed. According to the model, the multi-objective optimization proposition of integrated control performance and communication cost is proposed. The optimization of coupling structure will change the structure of coupling part of the original proposition, and by ignoring the coupling between some subsystems. In order to reduce the computational burden, the algorithm constructs an event trigger based on the stability of the system, when the trigger is triggered. Solving an optimization proposition including the decision and control performance of the coupling structure, updating the primary coupling structure, otherwise, keeping the coupling structure unchanged after the optimization, only solving the optimization proposition of the control performance. Compared with the distributed model predictive control algorithm based on dynamic coupling model, the algorithm reduces the computational complexity.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP273

【相似文献】