This paper is motivated by the basic need to develop methods for on-line detection of abnormal conditions in large electric power systems. In order to implement truly effective near-automated tools for this purpose, it is necessary to overcome several problems such as: (1) excessive computational complexity; (b) unacceptable approximations; and, (3) dependence on full state measurements. In an attempt to overcome these major roadblocks, we combine tools capable of producing accurate results over broad ranges of conditions, such as off-line data mining and machine learning, with the approximate, well-understood deterministic methods, such as sensitivity-based methods. The resulting approach indirectly overcomes the dependence on full state measurements; the actual choice of the most relevant measurements becomes a result of such a combined approach. The proposed approach is illustrated on an example of detecting a given voltage threshold violation.
[1]
J. Ross Quinlan,et al.
Induction of Decision Trees
,
1986,
Machine Learning.
[2]
J. C. Sabonnadiere,et al.
Structural analysis of the electrical system
,
1989
.
[3]
P. Lagonotte.
Probabilistic approach of voltage control based on structural aspect of power systems
,
1991
.
[4]
Simon Kasif,et al.
A System for Induction of Oblique Decision Trees
,
1994,
J. Artif. Intell. Res..
[5]
J. C. Sabonnadiere,et al.
Structural Analysis of the Electrical System: Application to Secondary Voltage Control in France
,
1989,
IEEE Power Engineering Review.