Prediction of DC load-rejection dynamic overvoltages using pattern recognition

Power system DC load-rejection may result in system dynamic overvoltages sometimes followed by generator self excitation. These overvoltages may be unacceptably high. The authors introduce a new pattern-recognition-based algorithm to predict power system load-rejection overvoltages. Using a power system digital simulation, a training set of a wide range of system loading conditions is created. A discriminant hyperplane is then derived to identify overvoltages, i.e. insecure conditions. In addition, a least squares method is incorporated to derive the relationship between the voltage security indices and postcontingency voltage deviations in the training process. A corrective algorithm based on the sensitivity analysis is also developed so that power system operators can take appropriate action at the appropriate time