Motivation:

Kalman filtering is an approach which is used in many control systems to provide a state estimate for a subsequent controller. It obtains the optimal solution (in the sense of an optimization problem) by combining predictions using a system model with the observations on hidden state variables measured by sensors. In a large system, the dimension of the state and the number of sensors will be large. Using a fusion center in the network combine all observations and perform centralized Kalman filtering will be impractical because of long communication distances and high computational effort. Moreover, it poses a single point of failure and latency which will also restrict the performance of centralized scheme. In this work, our motivation is to distribute the global system into low dimensional, inter-coupled sensor-based subsystems and design a distributed Kalman filter correspondingly.

Objective:

The main objective of this thesis is to:

1. Research and investigate methods from the state of the art in order to decompose the whole large-scale dynamic system into small subsystems;

2. investigate one distributed Kalman filter algorithm to perform state estimation locally at each subsystem with information exchange between subsystems;

3. achieve a consensus-based estimation result with respect to each state variable.

The student should firstly study how the a centralized Kalman filter works in a large-scale system. Then research on some data fusion methods including sensor network topology design and consensus approach will be needed with the purpose on reducing the communication requirement.