Convexification of bad data and topology error detection and identification problems in AC electric power systems

This study is motivated by major needs for accurate bad data detection and topology identification in the emerging electric energy systems. Due to the non-convex problem formulation, past methods usually reach a local optimum. This deficiency may lead to wrong bus/branch modelling and inappropriate noise assumption, causing significantly biased state estimate, incorrect system operation, and user cutoff. To overcome the local optimum issue, the authors propose in this study how to convexify bad data detection and topology identification problems to efficiently locate a global optimum result. To reduce relaxation error in the convexification procedure, a nuclear norm penalty is added to better approximate the original problems. Finally, they propose a new metric to evaluate the detection and identification results, which enables system operator to know how confidence one is for further system operations. Simulation results performed for several IEEE test systems show promising results for the future smart grid in improved accuracy.