项目名称: 交替方向乘子法研究及其在分布式信号与信息处理中的应用
项目编号: No.61471295
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 无线电电子学、电信技术
项目作者: 梁军利
作者单位: 西北工业大学
项目金额: 83万元
中文摘要: 本项目主要研究交替方向乘子法及其在分布式信号与信息处理中的应用。交替方向乘子法能够:1)对大规模问题实现等价分解及并行计算,最终通过局部问题求解获得大规模问题最优解;2)通过一致性约束使得传感器网络环境下各结点无须共享所获局部数据仅需交换局部解实现传感器网络环境下的分布式求解。然而该方法存在收敛速度慢、对应一致性约束的乘子多更新繁琐等缺陷,限制了其在传感器网络环境下的分布式计算及大数据计算方面的应用。为此,本项目主要研究自适应变步长更新以及乘子的简约进而改善算法的有效性。并在此基础上,研究用于多静态连续主动声纳的分布式检测前跟踪新方法、阵列信号分布式处理新方法、分布式存储数据的降维新方法。这些新理论的完善以及新算法的提出将为交替方向乘子法进一步的发展以及在涉及分布式信号与信息处理的多静态连续主动声纳、传感器网络、阵列信号处理、大数据分析与处理等领域内的应用提供重要的理论依据与技术支撑。
中文关键词: 交替方向乘子法;分布式信号与信息处理;传感器网络;阵列信号处理;降维
英文摘要: This project studies the alternating-direction method of Multiplier (ADMM) and its application to distributed signal and information processing. ADMM can: i)decompose the large-scale problem into equivalent local questions, and solve them in parallel to obtain the optimal solution to the original problem finally; and ii)enforce the consensus constraint on all nodes and exchange the local solutions rather than sharing the local data to solve the problem distributedly. However, the ADMM method converges slowly and updates so many Multipliers, which limits its application into Big Data computation and distributed computation in sensor networks. To solve these problems, this project will study the adaptive step size update and simplify the Multiplier update to improve its efficiency. Based on these improvements, this project will study the track-before-detect method for Multistatic continuous active sonar, distributed array signal processing mehtods, and dimensionality reduction methods for distributed stored Big Data. These new theory and methods will accelerate the development of ADMM, widen the application of ADMM, and support the development of Multistatic continuous active sonar, array signal processing, sensor network, analyses and processing of Big Data.
英文关键词: Alternating-direction method of Multiplier(ADMM);Distributed signal and information processing;Sensor network;Array signal processing;Dimensionality reduction