Research Highlights
Network Model Parameter Error Detection and Correction
By Dr. Ali Abur and Yuzhang Lin (PhD student)
Most applications in power system operation today rely on accurate network models. The network parame-
ter database, however, can be corrupted due to a number of reasons: inaccurate manufacturing data, human
data entry error, unreported device replacement and upgrade, operating status update failure, and ambient
condition variations. Based on state estimation, we develop a very effective methodology for the detection
and correction of parameter errors, which will facilitate the maintenance of a clean network parameter data-
base, and benefit a number of model-based applications in power system operation.
This method exploits the Lagrange multipliers associated with parameter errors when solving the state esti-
mation problem by the constrained weighted least squares method. It is applied jointly with the well-known
normalized residual test for measurement error identification, and is referred to as the Normalized Lagrange
Multiplier/Normalized Residual (NLMNR) test. Key features of this approach include: (1) Capability of differ-
entiating between parameter errors and measurement errors; (2) No need to make an a priori selection of
suspicious parameters; (3) No need to modify core state estimation software code.
Figure 1: CPU time of full
computation vs. proposed algorithm
Figure 2: Memory requirement of full
computation vs. proposed algorithm
By exploiting the sparse nature of power networks, a highly efficient implementation of the NLM test has
been developed. It avoids the full computation of a series of large dense matrices, and significantly reduces
the computational burden of this approach. Simulation results show that the required CPU time (Fig. 1) and
memory requirements (Fig. 2) will be very modest, even when this algorithm is executed on a very large real-
world power system (>14,000 buses).
Figure 3: Performance of the NLMNR test
for a parameter whose error is difficult to detect
5
CURENT Newsletter
Spring 2017
Figure 4: Performance of the NLMNR test
on an entire large utility system