A decentralized ID algorithm for detecting slow-fast oscillations in power systems from overwhelming volumes of phasor data

As the number of sensors, namely Phasor Measurement Units or PMUs, in the US power transmission grid scales up into the thousands within the next few years, the current state-of-the-art centralized data processing architecture will no longer be sustainable, and decentralized algorithms must be developed instead. In this paper we propose such an algorithm for one of the most critical applications in power system monitoring- namely, modal decomposition of swing dynamics for detecting slow and fast oscillation modes in the system with evaluation of their respective damping factors. Given a multiple set of coherent generation clusters in the system, we first use data from all PMU sources to calculate the oscillatory modes, their damping and participation in a centralized fashion. Next, we categorize the PMUs into several disjoint sets, and use the data from each of these sets to evaluate the modal frequencies for the entire system individually, assuming that the network has a connected topology guaranteeing system observability. A global estimate for any specific eigenvalue of interest is then computed from the geometric mean of those obtained from the disjoint estimation, and analytical expressions are derived to indicate how this geometric mean, representing the `fused distributed solution' compares to the centralized solution. A discussion on how the output nodes in the network should be chosen appropriately contingent on the topological structure of the network, in order to minimize the error between the two estimates is also presented. We illustrate our results with prototype power system network models inspired by two well-known transfer paths in the US west coast grid.