On-line prediction of transient stability using decision tree method — Sensitivity of accuracy of prediction to different uncertainties

The paper investigates sensitivity of accuracy of online prediction of power systems transient stability to different uncertainties. The prediction is based on measurements of generator angles and speeds during the post-fault condition provided by phasor measurement units (PMU), and application of Decision Tree (DT) data mining method. A DT is trained using large (16 machine, 68 bus) test system. Test datasets are designed to incorporate various uncertainties in the system, including fault duration and location, system operating condition and pre-fault system topology. The prediction accuracy is assessed at different times following the fault clearance. Only three phase faults, as the most critical ones for system stability, were used at this stage of research. It is demonstrated that DT can predict post-fault system state (stable/unstable) with over 88% accuracy as fast as 0.2 seconds following the fault. The accuracy of prediction typically increases to over 95% between 0.5 and 2.5 seconds following the fault depending on type of uncertainty involved.