Diversified parameter estimation in complex networks

Parameter estimation arises in the operation of many complex networks that are comprised of multiple interdependent sub-networks. Designing parameter estimators depends strongly on the extent of information available about the dynamics of network and the correlation structure among different parameters across the networks. Motivated by the core premise that designing state estimation models become more challenging as the networks grow in scale and complexity (primarily due to increasing interconnections in complex networks) identifying the best estimation model becomes increasingly challenging. By capitalizing on the measurements diversity in the complex networks, this paper proposes a learning-based framework for 1) dynamically identifying the best estimation model from a group of candidates for each subnetwork, and 2) aggregating the local estimates in order to form a globally optimal one. The analysis reveals that the framework is capable of providing a performance that has a diminishing gap with that of the best estimation model for each subnetwork without requiring any information about network dynamics.