Transient stability assessment based on artificial neural networks (ANNs) has shown much potential for online application in view of the computation efficiency of ANNs. However, two shortcomings of ANNs hinder them to be employed in an online environment. Firstly, with the increases of the system size, the training samples increase dramatically. The training burden, therefore, becomes too heavy. To this problem, a method for continuous attribute discretization based entropy is proposed to compress the training sample set in this thesis. Secondly, misclassifications in the boundary region between the two classes are in fact unavoidable due to the complexity of TSA input dimension and the limitation of ANN. To this problem, an assembling classification schemes is proposed by integrating different ANN classifiers based attribute matrix in rough set theory to improve the classification reliability. The application examples of the 10-machine 39-bus power system show the validity of the proposed methods for transient stability assessment.
|