Research On Consensus-based Kalman Filtering Algorithm And Its Application

Along with the wireless sensor network more and more widely applied in differentaspects of our daily social life, fusing the local information with noise and differentuncertainty reasonably and effectively is a important guarantee that wireless sensor networkcan acquire true information of the perception object. As a novel distributed informationfusion algorithm, Consensus-based Kalman do not need fusion center, the sensor nodes onlyneed to exchange the local data with its neighbor nodes, and all the nodes can obtainapproximate and high accuracy estimation value. Meanwhile, the algorithm’s convergencerate is fast and estimation performance is good, communication requirement is small. Wirelesssensor network’s applicability and robustness of different kinds of complex environment isimproved.This dissertation focuses on the different uncertainty of adjacency nodes’ estimate valuein wireless sensor networks and inaccuracy model or time-varying noises in practicalapplications. A quantization function used for quantizing the uncertainty of adjacency nodes’estimate value and a adaptive de-noising strategy is proposed used for improving thealgorithm’s estimation performance. Main contributions are structured as follows:Firstly, the wireless sensor network’s modeling method and the basic knowledge ofcomplex network are introduced; the Kalman filtering algorithm and consensus algorithminvolving in consensus-based Kalman filtering algorithm is also introduced, meanwhile somesimulations of correlation algorithm prepared for the following research is done.Secondly, aiming at the different uncertainty of adjacency nodes’ estimate value inwireless sensor network, considering the effect of node degree on estimate accuracy, aquantization function used for quantizing the uncertainty of adjacency nodes’ estimate valueis proposed. And then the quantitative value is used to optimize the consensus protocol as afusion weight of the adjacency nodes’ estimate value. The effectiveness of the novel algorithmis proved by simulations of mobile target tracking problem under the regular network, smallworld network and random network through change the uncertainty of the random selectednodes.Thirdly, As Consensus-based Kalman Filtering increases the estimation error of thealgorithm, owing to inaccuracy model or time-varying noises in practical applications, a noveladaptive de-noising Consensus-based Kalman Filtering algorithm is proposed basis ofanalyzing the effect of each part of the algorithm in filter process. The algorithm canadaptively adjust the observation noise covariance matrix, the optimal forgotten factor is usedfor calculating the observation noise covariance matrix, and the current observation data has the greater weight in the filtering process. Meanwhile, the algorithm is applied in mobiletarget tracking problem, the effectiveness of the novel algorithm is proved by analyzing theestimation error and the inconsistency estimation error.Finally, the adaptive de-noising Consensus-based Kalman Filtering is coupled withflocking control theory of multi-agent system. Assuming that each time the agent only canobtain the exact location information and velocity information of the leader by the means ofstate estimation, the agent plan itself motion of the next time based on the estimation value, amulti agent flocking control theory based on Consensus-based Kalman filtering is proposed.And the simulation result demonstrates that the multi agent system not only can realizeseparation, polymerization and speed matching, but also the speed of synchronization isimproved.