Power system security assessment and enhancement using artificial neural network

A power system is continually subjected to external and internal disturbances that are capable of causing instability in the system. The process of determining the stability of the system following the disturbances is known as security assessment. In particular, dynamic security assessment evaluates the stability of the power system with the time-dependent transition from pre-fault to post-fault states taken into consideration. For large disturbances, critical clearing time is a measure of the stability of the power system. The critical clearing time is a complex function of many variables, and its determination using conventional methods such as numerical integration is generally a time consuming and computationally intensive task. As an alternative approach, the artificial neural network is used in this paper to predict the critical clearing time. In particular, a multilayered feedforward neural network with error backpropagation algorithm was used to predict the critical clearing time of 2 different electric power systems; a 2 machine 5 bus system and a 3 machine 8 bus system. For the former power system, the optimal result of a percentage mean absolute error of 0.6% was obtained with a neural network structure of 1 hidden layer, 18 hidden neurons and the logistic activation function. The larger system had an optimal result of percentage mean absolute error of 2% with a neural network structure of 3 hidden layers, 30 hidden neurons and the logistic activation function.