Real-Time Identification of Electromechanical Oscillations via Deep Learning Enhanced Dynamic Mode Decomposition

The role of hydrogenerators to maintain the dynamic security by increasing the stability properties of the power system is required in the current energy transmission. This work proposes a reliable identification approach for poorly damped low-frequency power oscillations (LFO's). The applied method is a Dynamic Mode Decomposition (DMD) autoencoder neural networks for real-time identification and monitoring of electromechanical oscillations in the power system. An open-source Python library (ANDES) is used for modeling and generating training data for the autoencoder networks. Results show that a successful reconstruction of the oscillation mode can be made by combining both frequency and voltage measurements in the DMD approach. Moreover, the results show that DLDMD achieves a better fit of the underlying dynamics of the system than DMD.