ABSTRACT
This chapter seeks to resolve the problems associated with fully connected decentralized estimation: limited scalability, redundant communication, excessive computation and vulnerability to communication link loss. Two related techniques are proposed as a composite solution: model distribution and model defined internodal communication. Model distribution is the process of constructing reduced order models from a global system model by creating local state vectors which consist of locally relevant states such that there is dynamic equivalence between local and global models. A nodal transformation creating linear combination of states as local states does not give rise to redundant states, unless there are such states in the original system. This is because this derived transformation changes the basis and model size, but not the dynamic properties of the system. Several simplified expressions of the information space transformation matrix can be obtained by imposing constraints on the nodal transformation matrices, observation models and noises.
