Research On Distributed Estimation And Optimization Algorithm For Networked Systems

With the continuous development of computer networks, a variety of network systems spring out. Networked systems possess two characteristics. Firstly, the whole system can be divided into several subsystems, which communicates with each other through the network.Secondly, each individual node can cooperate with each other, and the whole network topology is distributed or semi-distributed. Most networks in our daily life can be seen as one of networked systems. Such as wireless sensor networks, mobile data networks, vehicle networking, industrial Internet, and so on.At present, many researches have been focused on the networked systems, especially the fully distributed networked systems. These researches focus on the study of distributed algorithms based on the characteristics of the network. In distributed algorithms, each node solves all kinds of problems produced by the networked systems only through the information interaction with its neighbors. Compared with the traditional centralized algorithm, in the distributed algorithm, there is no need for control node, and the status of each node in the network is equivalent. This can avoid the problem of single point failure and excessive load on key nodes. At the same time, it will increase bandwidth consumption and has limit to network topology.However, most of the researches on distributed algorithms for networked systems have been too ideal to be applied in the actual environment very well. Therefore, it is necessary to describe all kinds of actual scenes and establish corresponding models to reflect the problems in the actual environment. In this paper, we consider two scenes. One is dynamic wireless environment, in which links are not stable and the other is the quantization environment, in which, high precision data transmission is constrainted because of the limited bandwidth and resources. In addition,in order to demonstrate the advantages of distributed algorithms under large-scale network systems better, we also consider specific problemsand use the corresponding methods to solve. The contributions of this paper are as follows.(1) In virtual wireless sensor networked system, considering the uncertainty characteristics of link state, this paper proposes a dynamic link model. Based on this model, a dynamic diffusion Kalman filtering algorithm is proposed. The algorithm has a good performance in estimate tracking problem of moving objects in real-time system. In order to analyze the convergence and stability of the improved algorithm, the mean and mean-square error of the proposed algorithm are rigorously derived. On this basis, we conducte a simulation experiment of this algorithm. By comparing it with the existing algorithms, we can prove that even the algorithm with link probability of 0.1 still has a good stability and convergence performance, which is adapt to the application of networked systems in the dynamic network topology.(2) In the mobile crowd sensing network system, there are large number of distribution equipments with low power computing, storage resources, limited bandwidth, and so no, which will lead to the transmission problem of large-scale high-precision data. In this paper, we propose the quantification model based on random jitter. Then we presents quantitative diffusion Kalman filtering algorithm. In the algorithm, the information of the interaction between nodes is quantized, including observations, noise values, and so on. The quantized data will inevitably have a negative impact on the stability and convergence of the original algorithm. Therefore, we focus on the analysis of the performance of the quantization algorithm and verify it through simulation experiments.Through the analysis, we can see that the algorithm has good performance and can be applied to the networked system requirements in the quantization environment.(3) In the 5G content centric networked system, considering content caching strategy problems, a projection-based diffusion distributed optimization algorithm is adopted to solve the problem. In order to apply this algorithm, we first abstract the content caching policy and propose various assumptions so that we can apply our distributed optimization algorithm. In the construction of objective problems, the whole network energy consumption is taken into account. Then, it focuses on how the distributed optimization algorithm solves the problem. Through experiments, the problem is simulated in the actual scene, and achieves the desired results.It is worth nothing that, there are a variety of networked systems and distributed algorithms, all algorithms and applications proposed in the paper are to provide ideas and methods for the study of these problems.Therefore, our research is of great significant.