Reliable Bad Data Processing for Real-Time State Estimation

Abstract

The weighted least squares performance index test (J(x) test) conventionally used in power system static state estimation has poor reliability for detecting the presence of measurement errors in the range 3 to 20 standard deviations. This paper describes a very simple alternative method, with considerably improved bad data detection properties, based on evaluating the coherency between the measurement with the largest normalized residual and the remainder of the measurement system. In fact, the detection and identification phases of bad data processing become combined. Inanexistent state estimator using normalized residuals, the cost of implementation of the new method is negligible. The new method has been tested extensively on several power systems, and compared with the J(x) test.