The Analysis of State Estimator Condition Number: A Machine Learning Approach

Power system state estimation could be occasionally ill-conditioned because of the network structure and measurement types. This paper investigates the impact of different types of measurements, as well as the measurement placement, to the value of condition number. By using the machine learning approach, the implicit correlations between the condition number and the measurements can be revealed. Due to the paucity of the training data, we manually calculate the accuracy-critical points to form a proper training dataset. Further, not only the condition number itself, but also its liability is presented by using the Bayesian confidence probability. The proposed method is verified by the IEEE 34-bus and IEEE 123-bus feeder testbeds.