A hybrid approach for security evaluation and preventive control of power systems

This paper presents a hybrid approach for online security evaluation and preventive control of power systems. The artificial neural network (ANN) offers potential advantages regarding efficient computation and ease of knowledge acquisition. However it is a "black box" type approach, which lacks interpretability. The decision tree (DT) approach is known for its interpretability but comparatively less accurate. The proposed hybrid approach combines ANN and DT approaches to exploit their potential while suppressing their drawbacks. It applies an ANN for security evaluation of power systems and DT methodology to drive preventive control measures. A divergence based feature selection algorithm has been investigated to select an optimal combination of neural training features. The method has been applied on an IEEE power system and the results obtained are promising.