SOCP Convex Relaxation-Based Simultaneous State Estimation and Bad Data Identification

TL;DR

This paper introduces a convex optimization approach using SOCP relaxation for simultaneous power system state estimation and bad data detection, overcoming limitations of traditional residual-based methods.

Contribution

It proposes a novel joint estimation and bad data identification method based on SOCP convex relaxation and L1-norm residuals, enabling detection of complex bad data scenarios.

Findings

Effective detection of leverage point bad data

Simultaneous state estimation and bad data identification in large networks

Outperforms traditional residual-based methods

Abstract

Traditional largest normalize residual (LNR) test for bad data identification relies on state estimation residuals and thus can only be implemented after running Power System State Estimation (PSSE). LNR may fail to detect bad data in leverage point measurements and multiple interacting and conforming bad data. This paper proposes an optimization problem formulation for joint state estimation and bad data identification based on second-order cone programming (SOCP) convex relaxation. L1 -norm of the sparse residuals is added in the objective function of the state estimation problem in order to recover both bad data and states simultaneously. To solve the optimization problem in polynomial time, first, SOCP convex relaxation is applied to make the problem convex. Second, least squares error (LSE)-based semidefinite programming (SDP) cutting plane method is implemented to strengthen the SOCP relaxation of PSSE. Numerical results for the proposed algorithm demonstrate simultaneous state estimation and bad data identification for networks with small sizes and thousand buses. Compared to the LNR method, the proposed method can detect bad data in leverage point measurements and multiple conforming bad data.