Application of computational intelligence to power system security assessment

This dissertation investigates the potential of computational intelligence techniques applied to power system security assessment. The objective is to utilize computational intelligence, including neural networks and evolutionary algorithms, to identify the exact location of the power system security boundary relative to a given system operating point. Such information is potentially very useful in the everyday operation of the power system in that it could be used to avoid insecure operating regions or to regain security once it's lost. Neural networks offer several key advantages over traditional solution methods including the ability to learn from examples and the ability to extract information about the underlying system through neural network inversion. In this dissertation, multi-layer perceptron neural networks are trained to identify the power system security boundary given a set of predisturbance power system features. Evolutionary algorithms, coupled with a constrained gradient descent algorithm are then used to invert the neural network and identify specific areas of the security boundary. This dissertation also includes a review of the state of the art in applying neural networks to power system security assessment. Several methods of improving neural network performance are presented including techniques for increasing the training speed, improving the generalization ability and eliminating large errors during training. A review of neural network inversion techniques is presented. Fisher's linear discriminant, coupled with feature selection techniques is proposed as a means for reducing the dimensionality of the input space by selecting a subset of features for neural network training. A statistical analysis is performed to show the effectiveness of the Fisher feature selection technique. The affect of changes in system topology on the accuracy of the trained neural network is investigated and suggestions are given for minimizing this effect. Each of the proposed concepts are validated through simulations of realistic power systems including the IEEE 17-generator test system and the IEEE 50-generator test system.