Assessment of a Multi-Step LSTM-based Ensemble Strategy for Short-term Grid Modal Parameters Forecast

Abstract

The phenomenon of electromechanical Inter-Area Oscillation is becoming day-by-day a harder problem to face in modern power grid operation. The possibility to identify some time in advance possible critical situations is certainly a nice feature for cutting-edge power grid monitoring systems. In this context, the length of the prediction horizon constitutes a key factor, so, longer lengths translate directly to possibly more effective control actions. This paper presents an attempt to forecast the values of the modal parameters related to inter-area oscillations over extended time periods by using a multi-step prediction strategy integrating long-short-term memory units and ensemble methods. The building steps of the overall proposed approach are illustrated together with some preliminary results coming from the application of the method to real measurement data.