Structural identification of static-security assessment problems using self-organising and associative memory

Abstract

Security assessment is the process of continuous monitoring and analysis of contingencies to ensure that power systems are operated in a proper and secure manner. This paper presents an artificial neural network (ANN) based approach for power system static security assessment. The approach begins with a procedure of ANN based busbar voltage predictions. Using the prediction results, potential voltage problems, overloads and remedial measures are highlighted. The ANN is trained using a set of input-output data obtained from off-line analysis of power networks and SCADA measurements. The training time can be prohibitively long due to the huge size of training data used to ensure prediction accuracy. In order to relieve the training burden, this paper proposes the use of a reduced training data set. A self- organising map (SOM) is used at the front-end to map a high-dimensioned space of input-output data into a low-dimensioned training data set. The mapping identifies the functional structure of the study power system, preserves the non-linear topological characteristics of the original data space, and therefore provides a global representation of the systems voltage/ control characteristics. Results from a medium- size study system are described in this paper to demonstrate the effectiveness and salient features of the SOM as a means of data reduction for ANN based security assessment.