Power System State Estimation and Bad Data Detection by Means of Conic Relaxation

Abstract

This paper is concerned with the power system state estimation problem, which aims to find the unknown operating point of a power network based on a set of available measurements. We design a penalized semidefinite programming (SDP) relaxation whose objective function consists of a surrogate for rank and an l1-norm penalty accounting for noise. Although the proposed method does not rely on initialization, its performance can be improved in presence of an initial guess for the solution. First, a sufficient condition is derived with respect to the closeness of the initial guess to the true solution to guarantee the success of the penalized SDP relaxation in the noiseless case. Second, we show that a limited number of incorrect measurements with arbitrary values have no effect on the recovery of the true solution. Furthermore, we develop a bound for the accuracy of the estimation in the case where a limited number of measurements are corrupted with arbitrarily large values and the remaining measurements are perturbed with modest noise values. The proposed technique is demonstrated on a large-scale 1354-bus European system.