Power System Security Boundary Enhancement Using Evolution–

This paper presents a new method for enhancing the accuracy of partially trained multilayer perceptron neural networks in specific operating regions. The technique is an extension of previously published query learning algorithms and uses an evolutionary-based boundary marking algorithm to evenly spread points on a contour of interest in this case the power system security boundary. These points are then presented to an oracle (i.e. simulator) for validation. Any points that are discovered to have excessive error are then added to the neural network training data and the network is retrained. This technique has advantage over existing training methods because it produces training data in regions that are poorly learned and thus can be used to improve the accuracy of the neural network in these specific regions. An example of the proposed algorithm is applied to the IEEE 17 generator test system.