A SIPSS-Lasso-BPNN scheme for online voltage stability assessment

Load active power at the voltage collapse point (P_LL) is a useful index for online voltage stability assessment. This paper proposes a SIPSS-Lasso-BPNN scheme to offline fitting and online forecasting the load active power at the voltage collapse point P_LL. The scheme consists of a SIPSS (Similarity Index of Power System State) based screening method, a Lasso (Least absolute shrinkage and select operator) method and a back propagation neural network (BPNN). The SIPSS based screening method screens the training samples according to their similarity indexes of power system states. The Lasso method selects the principal input features which are most explanatory to P_LL via the shrunken regression analysis. The training samples are reduced by the above two methods. The BPNN is used to offline fit and online forecast the P_LL through the reduced training samples. The test results on the New England 39 bus system shows that the SIPSS-Lasso-BPNN scheme can significantly improve the efficiency of BPNN offline training and guarantee the forecasting accuracy.