On-line Inter-area Oscillatory Stability Assessment with Credibility for Power Systems Based on Exploring Connotative Relationships in Massive Data

摘要

An integrated approach for on-line oscillatory stability assessment (OSA) based on exploring connotative relationships in massive data is proposed, which consists of three stages and can give credible OSA results. The approach has performed higher accuracy than some conventional techniques by avoiding the use of potential inaccurate assessments. The approach is a kind of transparent tool, which can provide a clearer relationship between the operation variables and the onset of an instability event than black-box tools for system operators. Compared with the conventional transparent tool decision tree, the approach is more preferable in some aspects: good robustness to the data missing of partial input features; searching surrogates for missing features easily, and avoiding tedious debugging. Moreover, the approach considers both classification and prediction for application, and the credible decision-making rules are presented. In addition, the approach can accommodate the on-line variation of system operation condition brought by different factors in practical application. In the approach, the relationships between operation variables and oscillatory stability margin are assigned scores by the maximal information coefficient and the Pearson correlation coefficient. The connotative non-linear functional relationships and linear ones are explored by ranking the scores, and some top relationships are shown and curve fitted. A processing unit for OSA is designed based on the fitted top relationships in the approach. The performance is examined on the IEEE 39-bus test system and a practical 1648-bus system provided by the software PSS/E. The impacts of training set size, selected relationships' total number, and types on the accuracy are studied. The robustness of the approach to variation of topology, distribution among generators/loads, and peak load/minimum load is analyzed.