Intelligent Distributed Estimation Architectures

智能设备和日常用品已经配备了多个传感器，产生了大量的传感器数据，这些数据需要过滤、分析、监控和处理。为了克服这一挑战，通常将大量传感器数据划分为子集，并将其馈送到并行运行的多个滤波器中。这种方法可以理解为自底向上的方法，它依赖于多个独立的局部过滤器的设计和运行，收集估计结果，并将它们融合到全局估计中。由于每个局部滤波器是单独设计的，并且不知道全局估计目标，自下而上的方法通常不能产生最优的和计算效率高的全局估计。该项目通过主要关注所需的全局估计质量，从不同的角度来看待大规模分布式估计问题的解决方案。提出的自顶向下方法从全局估计质量的要求出发，设计局部滤波器的结构和拓扑以满足要求，同时尽量降低计算复杂度。考虑到分布式计算体系结构的最新发展，自顶向下的方法尤其重要，分布式计算体系结构为用户和设计人员提供了巨大的灵活性和可变性，以实现不同的和可能随时间变化的评估目标。该项目将是一个新颖的尝试，因为它主动地根据给定的全局评估目标来改变评估体系结构。使用这种方法，可以建议对体系结构进行调整，而不必对给定的体系结构进行广泛的试验。为此目的，项目将分析评估体系结构的构建块；即数据流、网络拓扑、融合策略和过滤参数，重点关注它们对全局目标的个人和合作影响。体系结构中的处理单元还具有监视和故障检测功能，以便使体系结构适应不断变化的操作条件和需求。我们认为，在物联网时代，这是一个及时的考虑。

Smart devices and everyday items equipped with a plurality of sensors - already today - generate massive amounts of sensor data that are to be filtered, analyzed, monitored, and processed. To overcome this challenge, abundant sensor data are typically partitioned into subsets and fed into multiple filters running in parallel. Such an approach, which can be understood as the bottom-up approach, relies on a design and run of multiple independent local filters, collecting the estimation results, and fusing them into a global estimate. As each local filter is designed separately and is not aware of the global estimation goal, the bottom-up approach cannot generally result in an optimal and computationally efficient global estimate.This project views the solution to the large-scale distributed estimation problem from a different perspective by primarily focusing on the desired global estimation quality. The proposed top-down approach starts from the requirements on the global estimation quality and designs the structure and topology of local filters to meet the requirements while the computational complexity is kept as low as possible. The top-down approach is especially important considering the recent developments in distributed computing architectures which give an enormous flexibility and variability to the users and designers to achieve different and possibly time-varying estimation goals.The project will be a novel attempt due to its proactive approach to change the estimation architecture with respect to a given global estimation aim. With such an approach, it is possible to suggest an adaptation to the architecture without doing extensive trials with a given architecture. For this aim, the project will analyze the building blocks of the estimation architecture; that is the data flow, network topology, fusion strategy, and filter parameters with a focus on their individual and cooperative impact on the global objective. The processing units within the architecture also feature monitoring and fault detection functionalities in order to adapt the architecture to changing operational conditions and requirements. We believe this to be a timely consideration in the age of the Internet of Things.