-
分布式凸优化问题是指在缺乏全局优化问题的完整信息的情况下,系统中的多智能体协作地求解全局代价函数的最优解.这类问题的解决需要设计完全分布式的优化算法,也就是优化算法由没有中央协调器的智能体来实现.分布式优化问题最近取得了许多重要成果[1-12],基于一致的分布式策略正成为解决优化问题的主流.这已经产生了许多优化算法,包括分布式次梯度下降算法[1]、分布式对偶平均算法[2]、分布式nesterov梯度算法[3]等.随后,这一系列工作被扩展到各种现实条件下的分布式优化,如随机次梯度误差[4]、随机通信网络[5]等.实际上,上述分布式优化算法[2-5]的实现都需要构建一系列双随机矩阵,阻碍了这些算法的开发和实际应用,特别是在随时间变化的一般非平衡有向实际网络环境中,因为双随机矩阵的条件难以以分布式方式满足.为了解决上述非平衡问题,最近的文献[6-8]分别从不同的角度解决非平衡性.值得注意的是,尽管文献[6-8]中的算法避免了构造双随机矩阵,但它们都要求所有智能体准确地知道其入度邻居的出度信息进而必须构造列随机权重矩阵.然而,构造列随机矩阵在某些方案(例如基于广播的通信方案)中可能是不现实的.
当网络是非平衡有向且相应的权重矩阵是行随机时,Mai和Abed[9]提出了一种改进的分布式次梯度投影方法,用于求解具有全局约束集的分布式优化问题.在代价函数是凸且光滑的情况下,理论分析证明了文献[9]中的算法渐进收敛于优化问题的最优解.遗憾的是,文献[9]中的算法不能解决带有耦合线性约束和局部不等式约束的分布式优化问题.因此,Doan和Beck[10]开发了一种分布式拉格朗日网络资源分配方法,该方法能够很好解决带有耦合线性约束和局部不等式约束的分布式优化问题.但是,在文献[10]中,网络拓扑结构是无向的,这在实际网络环境中是不切实际的.
综上所述,在非平衡有向网络上研究带有耦合线性约束和局部不等式约束的分布式优化问题并将其应用于电力系统中分布式经济分配上将会是一项非常有意义的工作.因此,本文的主要贡献可以归纳为以下3个方面:①在非平衡有向网络上提出一种新颖的分布式原始-对偶次梯度算法解决带有耦合线性约束和局部不等式约束的分布式优化问题,与文献[10]相比,本文算法适用于非平衡有向网络;②虽然文献[9]同样考虑了行随机矩阵,但是本文算法能够处理带有耦合线性约束和局部不等式约束的分布式优化问题;③本文算法最终渐进收敛到全局优化问题的最优解.
A Primal-Dual Algorithm for Solving Distributed Economic Allocation Problem Over a Directed Unbalanced Network
-
摘要: 受电力系统经济分配问题的启发,研究了分布式经济分配问题,其主要目标是在m个智能体组成的非平衡有向网络上最小化m个局部凸代价函数之和.网络中的每个智能体都仅仅知道自己私有的局部凸代价函数,并且同时受到耦合线性约束和局部不等式约束的影响.此外,特别关注每个智能体仅允许通过不平衡有向网络与其内部邻居进行交互的情况.为了分布式地解决上述问题,提出一种新的只需要智能体进行本地计算和本地通信的完全分布式原始-对偶次梯度算法.当网络拓扑是强连通的且权重矩阵是行随机时,理论分析证明本文的算法可以渐进收敛到全局优化问题的最优解.最后,给出了电力系统中分布式经济分配问题的数值仿真,验证了所提出算法的有效性和分析过程的正确性.Abstract: Inspired by the economic allocation problem in power systems, this paper studies the distributed economic allocation problem in power systems where the main goal is to minimize a sum of local convex cost functions over a directed unbalanced network composed of agents. Each agent in the network privately knows its own local convex cost function and is subjected to both coupling linear constraint and individual inequality constraints. Moreover, we particularly focus on the scenario where each agent is only allowed to interact with its in-neighbors over a directed unbalanced network. In order to solve the above problems distributedly, we propose a new fully distributed primal-dual subgradient algorithm that only requires the agent to perform local computing and local communication. Most of the existing algorithms require all agents to possess the out-degree information of their in-neighbors, which is impractical and hardly inevitable as interpreted in the paper. When the network topology is strongly connected and the weight matrix is row stochastic, theoretical analysis proves that our algorithm can converge to the optimal solution of the global optimization problem. Finally, we present a numerical simulation of the distributed economic allocation problem in power systems to verify the effectiveness of the proposed algorithm and the correctness of the analysis process.
-
[1] doi: http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=de6cc5daca0f6091a2c881d683515383 NEDIC A, OZDAGLAR A. Distributed Subgradient Methods for Multi-Agent Optimization[J]. IEEE Transactions on Automatic Control, 2009, 54(1):48-61. [2] DUCHI J C, AGARWAL A, WAINWRIGHT M J. Dual Averaging for Distributed Optimization:Convergence Analysis and Network Scaling[J]. IEEE Transactions on Automatic Control, 2012, 57(3):592-606. doi: 10.1109/TAC.2011.2161027 [3] CHEN A I, OZDAGLAR A. A Fast Distributed Proximal-Gradient Method[C]//201250th Annual Allerton Conference on Communication, Control, and Computing (Allerton). Monticello: IEEE Press, 2012: 601-608. [4] SUNDHAR RAM S, NEDIC A, VEERAVALLI V V. Distributed Stochastic Subgradient Projection Algorithms for Convex Optimization[J]. Journal of Optimization Theory and Applications, 2010, 147(3):516-545. doi: 10.1007/s10957-010-9737-7 [5] MATEI I, BARAS J S. Performance Evaluation of the Consensus-Based Distributed Subgradient Method Under Random Communication Topologies[J]. IEEE Journal of Selected Topics in Signal Processing, 2011, 5(4):754-771. doi: 10.1109/JSTSP.2011.2120593 [6] doi: http://d.old.wanfangdata.com.cn/NSTLQK/NSTL_QKJJ0234916548/ NEDIC A, OLSHEVSKY A. Distributed Optimization over Time-Varying Directed Graphs[J]. IEEE Transactions on Automatic Control, 2015, 60(3):601-615. [7] 张豪, 韩易言, 吕庆国, 等.时变网络拓扑图下智能电网中基于优化算法的分布式调度响应[J].西南大学学报(自然科学版), 2018, 40(7):177-180. doi: http://xbgjxt.swu.edu.cn/article/doi/10.13718/j.cnki.xdzk.2018.07.024 [8] XI C, KHAN U A. DEXTRA:A Fast Algorithm for Optimization Over Directed Graphs[J]. IEEE Transactions on Automatic Control, 2017, 62(10):4980-4993. doi: 10.1109/TAC.2017.2672698 [9] MAI V S, ABED E H. Distributed Optimization Over Weighted Directed Graphs Using Row Stochastic Matrix[C]//2016 American Control Conference. Boston, MA: IEEE Press, 2016: 7165-7170. [10] DOAN T T, BECK C L. Distributed Lagrangian Methods for Network Resource Allocation[C]//2017 IEEE Conference on Control Technology and Applications. Mauna Lani: IEEE Press, 2017: 650-655. [11] 张文光, 陈俊, 姚钰辉, 等.分布式网络环境中基于MapReduce的WordCount实现[J].贵州师范大学学报(自然科学版), 2015, 33(1):93-97. doi: 10.3969/j.issn.1004-5570.2015.01.020 [12] KAR S, HUG G. Distributed Robust Economic Dispatch in Power Systems: A Consensus + Innovations Approach[C]//Proceedings of 2012 IEEE Power and Energy Society. General Meeting. New York: IEEE Computer Society Press, 2012: 1-8.