A highly efficient bad data identification approach for very large scale power systems

The well-known largest normalized residual (LNR) test for bad data identification becomes computationally inefficient for large-scale power systems containing a large volume of bad data, given the fact that it identifies and removes bad measurements sequentially, one at a time. In this paper, a highly efficient alternative implementation of the LNR test will be presented where the computational efficiency will be significantly improved. The main idea is based on the classification of suspect measurements into groups, which have negligible interaction. Then, the LNR test can be applied simultaneously to each individual group, allowing simultaneous identification of multiple bad data in different groups. Consequently, the number of identification/correction cycles for processing a large volume of bad data will be significantly reduced. Simulations carried out on a large utility system show drastic reductions in the CPU time for bad data processing while maintaining highly accurate results. This work is expected to facilitate implementation and more effective use of the LNR test for identifying and correcting measurement errors in very large power systems.