| Abstrakt: |
Generator coherency identification is an important step that needs to be taken in the case of a fault to enable durable islanding that could prevent a cascading blackout. Coherent generators are generally identified using the behaviour of rotor angles of the generators. Deep learning has been an up-and-coming technique to solve complex problems with supreme accuracy. This research aims to implement deep learning into the problem of generator coherency identification, to determine coherent generator groups with extreme accuracy. For this purpose, LSTM was chosen as the deep learning technique to be implemented in this research and was fine-tuned to yield the most accurate results for this problem. Data was collected in the form of rotor angles using a simulation of the IEEE 39-bus test power system in DIgSILENT PowerFactory. Based on the results obtained, LSTM proved to be extremely accurate in the primary performance metric RMSE, with the lowest value of 0.03. However, in the secondary performance metric, R2, the value was lower at only around 0.81. However, in various tests such as with reduced input data of 1000, 500, 250 and 50 out of the original 2000, and reduced computational power of 50% and 90%, LSTM reigned accurate, in both RMSE and R2. A third performance metric, accuracy as a percentage, was also analysed, where LSTM had an accuracy of 95.7%, which was compared to an ANN model in the literature that had a 95.03% accuracy. This concludes that LSTM is applicable for the clustering of coherent generators. All in all, deep learning was able to yield more accurate results compared to traditionally used machine learning techniques such as ANN in literature. [ABSTRACT FROM AUTHOR] |
