A computational intelligence method to monitor voltage collapse

The paper explores the potential of artificial neural neural networks (ANNs) for the online monitoring of modern power systems with a view to identifying possible voltage-collapse scenarios and providing an advisory tool to a control engineer. Many conventional techniques for monitoring voltage-collapse are too computationally intensive to achieve real-time operation, hence the promise of ANNs, which may take enormous CPU time to train offline but, once trained, can operate extremely fast online. The major drawback with the ANN approach is the huge cardinality of the training set required to represent the input space sufficiently for the ANN to learn its input-output mapping accurately. The dimensionality problem can be tackled by partitioning a power system into physical subnetworks (e.g. “reactive areas” or “voltage zones”), followed by decomposition of each input space into several classes, with each subspace monitored by a separate ANN. Results are presented from studies on the Ward-Hale test system as well as a 40-node system