ANALYSIS OF BAD DATA IN POWER SYSTEM STATE ESTIMATION UNDER NON-GAUSSIAN MEASUREMENT NOISE

Abstract

In practice, aging power system measurement equipment may deviate the noise statistics of errors associated with measurements provided by a supervisory control and data acquisition system from a Gaussian distribution. Furthermore, recent research conducted by the Pacific Northwest National Laboratory in the United States reveals that errors associated with phasor measurement units do not obey a Gaussian probability distribution. Under these conditions, the estimated state of an electric power system may deviate from the actual operation state because of the generally adopted assumption that measurements’ errors follow a Gaussian probability distribution. Hence, the question of what effect non-Gaussian data have on the outcome of state estimators and of bad data analyses used in power system control centers must be investigated. This paper addresses this question by providing a comprehensive analysis of the quality of estimated states using both the weighted least squares method and the largest normalized residual test, which are the ones currently used in todays' control centers, when the error distribution of measurements is not truly Gaussian. In this context, this paper also aims to provide a thorough discussion of the suitability of applying a recently proposed bad data analysis, which is based on the largest normalized estimated residual test (LNERT), for detecting, identifying and correcting or eliminating multiple measurements with gross errors under Gaussian and non-Gaussian noises. The effectiveness of the proposed LNERT-based bad data analysis is extensively compared with respect to other proposals for bad data identification by means of state estimation studies performed in the IEEE 14-bus and 118-bus test systems as well as on the very large 7659-bus representation of the Mexican interconnected power system.