Rule extraction for voltage security margin estimation

Research efforts have been devoted to estimating voltage security margins which show how close the current operating point of a power system is to a voltage collapse point as assessment of voltage security. One main disadvantage of these techniques is that they require large computations, therefore, they are not efficient for on-line use in power control centers. In this paper, we propose a technique based on hyperrectangular composite neural networks (HRCNNs) and fuzzy hyperrectangular composite neural networks (FHRCNNs) for voltage security margin estimation. The technique provides us with much faster assessments of voltage security than conventional techniques. The two classes of HRCNNs and FHRCNNs integrate the paradigm of neural networks with the rule-based approach, rendering them more useful than either. The values of the network parameters, after sufficient training, can be utilized to generate crisp or fuzzy rules on the basis of preselected meaningful features. Extracted rules are helpful to explain the whole assessment procedure so the assessments are more capable of being trusted. In addition, the power system operators or corresponding experts can delete unimportant features or add some additional features to improve the performance and computational efficiency based on the evaluation of the extracted rules. The proposed technique was tested on 3000 simulated data randomly generated from operating conditions on the IEEE 30-bus system to indicate its high efficiency.