Sparse Estimation of Faults by Compressed Sensing with Structural Constraints

This paper addresses the challenge of fault location in large power networks using a limited number of sensors. It was recently shown that power system faults may be modeled by sparse vectors, and hence, can be located efficiently using sparse recovery techniques. In this paper, we extend this approach and propose a sparse recovery algorithm that exploits both the sparsity constraints and additional structural constraints imposed by the faults physical models. To this end, faults are represented by sparse vectors that are subjected to nonconvex constraints. These constraints are shown to provide additional information that is exploited to reduce the number of measurements and to improve the location accuracy. The algorithm searches directly over these physical faults, and therefore, operates over a small solution space. Simulations on the IEEE 118 bus test-case network show that 4 to 20 sensors are sufficient to recover faults with adequate accuracy. With 20 PMU sensors, more than 99% of single-fault events and 84%-95% of two-fault events are located, depending on fault types and SNR.