Partial state estimation for electricity grids

Power system operators rely critically on state estimation for verification, fault detection and localization, and re-dispatch under contingency operations. In current practice, power system data within a control area such as voltages, phases, real and reactive power flows and injections, are relayed to the operator using SCADA systems. State estimation is formulated as an over-determined weighted nonlinear least squares problem and the solver of choice is the Newton-Raphson method. Two critical issues are: (a) estimate quality, due to data latency or convergence to false local minima, and (b) computation time, due to the large number of state variables involved. In this paper, we explore techniques to accelerate state estimation by computing state estimates at a small subset of buses using limited measurements from the power subsystem of interest. These could be operator selected “important” buses which connect to “important” lines with significant real power transfer. Our techniques are inspired by uncertainty quantification methods. The influence of power flows from exogenous buses is treated as uncertainty which defines a feasible set of state variables consistent with available measurements. The state estimation problem can be cast as a non-convex optimization problem. We use a surrogate model relaxation and Shor´s rank relaxation to obtain state estimates and associated error bounds at user-defined confidence levels.