You are here

ECE PhD Defense: "Reliable and Efficient Methods for Identification of Parameter and Measurement Errors in Power Networks," Yuzhang Lin


442 Dana

August 13, 2018 11:00 am
August 13, 2018 11:00 am

The detection, identification, and correction of parameter and measurement errors (referred to as "error processing" below) are one of the core problems in the modeling and monitoring of electric power networks. In this dissertation, analyses and methodologies are proposed in order to tackle different aspects of this problem.

To start with, the theoretical foundation of error processing is strengthened. The fundamental relations between the errors and their corresponding indicators (Lagrange multipliers and residuals) which are provided in the output of SE, are derived. Along with the development of this result, several remaining theoretical issues regarding error processing are addressed, including formulation of the identification approaches as hypothesis testing procedures, proof of the "largest normalized variable" criterion, and detectability and identifiability of the errors.

Closely following the aforementioned results, avenues to enhance the power of parameter error detection and estimation methods are presented. They are aimed at reliably processing the "weakly detectable errors" which are likely to be missed by existing approaches. The proposed solutions include exploitation of multiple measurement scans, and strategic use of phasor measurement units (PMUs).

A  critical consideration in real-world implementation, namely the computational efficiency of error processing approaches is subsequently addressed in this dissertation. Solutions are presented to tackle the computational bottlenecks at different stages of error processing. They include efficient computation of the covariance matrix associated with Lagrange multipliers, linear correction of parameter errors, and simultaneous identification of multiple non-interacting measurement errors.

Other than investigating and enhancing the error processing approaches based on the output of conventional state estimation methods, this dissertation also seeks another avenue to process the errors by directly using robust state estimation. A novel approach named Extended Least Absolute Value State Estimator (ELAV SE) is proposed for this purpose. It is the first state estimation approach for power systems that remains robust in the presence of both parameter and measurement errors, while automatically identifying and estimating these errors at the same time.

Finally, the dissertation considers some of the cyber-security issues related to power system modeling. Based on the earlier studies in this dissertation, it is shown that by strategically injecting errors into the model parameter database, certain network applications involving the operation of electricity markets can be manipulated, while detection of such attacks will be highly unlikely. The criticality of ensuring the accuracy and security of the parameter database is thus illustrated for a very practical example of electricity markets.

  • Professor Ali Abur (Advisor)
  • Professor Bradley Lehman
  • Professor Hanoch Lev-Ari