Enhancement of short-term prediction capabilities of inter-area grid oscillations with a multi-variate ensemble-based method

The actual and future even higher penetration of renewable energy sources into the power grid sets challenging issues for transmission system operators, especially concerning the hard-to-solve problem of inter-area electromechanical oscillations. Despite the useful existing monitoring systems, the possibility of having predictive monitoring features for such phenomena could be an appealing tool. The work presented in this paper aims to assess the possibility of enhancing the predictive monitoring features offered by machine learning techniques based on the combination of ensemble methods and Long-Short-Term Memory units using multi-variate methods. The development steps of a multi-variate prediction strategy are presented together with the assessment of its performance versus uni-variate solutions. The assessment takes into account different kinds of datasets, taken from real grid measurements, and strategy configurations. Either transient low frequency oscillation phenomena or normal grid operation are considered as representative cases of real-world scenarios. Finally, some preliminary considerations about improving prediction performance and the limitations are outlined.