Comparison of LSTM-based Prediction strategies for Grid Modal Parameters Forecast

The high penetration of renewable energy sources poses great challenges for transmission system operators, especially concerning the detrimental phenomenon of electromechanical Inter-Area Oscillations. Although the actual monitoring techniques can offer a useful baseline in order to fight against such phenomena, predictive features are highly desirable in this context. This work presents a preliminary comparative study of two prediction strategies suitable to forecast the short-term values of the grid modal parameters. The considered strategies are based on the proper integration of the Dynamic Mode Decomposition technique with Machine Learning techniques such as Long-Short-Term Memory units and Ensemble methods. The development steps of both techniques are fully illustrated and the performance comparison is done by accounting for some key performance indicators. Two assessment scenarios are considered, based on the availability of some real measurement data.