ANN-based pattern recognition technique for power system security assessment

Abstract

Security assessment is to predict a power system's ability to withstand a set of next contingencies. An ANN-based pattern recognition method is used to perform static security assessment for power systems due to its potential in terms of speed and accuracy for online application. With the input pattern for ANN be composed of power system pre-contingency state described in busbar power injections (P, Q), the output pattern of ANN is composed of the performance index (PI) values of power system post-contingency state to a list of next contingencies. So the output vectors of ANN will indicate not only either 'secure' or 'insecure' state of the current system but also the severity of security limit violations under contingencies. To cope with the curse of dimensionality and improve efficiency of ANN, R-ReliefF algorithm is introduced to extract those variables that are with more discriminatory information from (P, Q) set to realise the nonlinear mapping from input space to output space. The proposed algorithm is tested on a 77-busbar practical power system with promising results.