VILNIUS TECH Library invites you to follow the published new dissertations. The dissertation „Investigation of recurrent neural networks-based methods for early fault detection and short-term power forecasting in wind energy applications“ prepared at VILNIUS TECH by Mindaugas Jankauskas. The dissertation was prepared in 2021–2026. Scientific consultant – Prof. Dr Artūras Serackis.
The dissertation was defended at the public meeting of the Dissertation Defence Council of the Scientific Field of Electrical and Electronic Engineering in the Aula Doctoralis Meeting Hall of Vilnius Gediminas Technical University at 10 a. m. on 5 June 2026.
The increasing role of wind energy in modern power systems creates a growing need for reliable turbine operation, accurate short-term power forecasting, and computationally efficient data-driven methods. This dissertation addresses two related problems: early fault detection in wind turbines using supervisory control and data acquisition (SCADA) time-series data, and short-term wind farm power forecasting using meteorological forecasts. The dissertation aims to develop and investigate data-driven methods that improve the accuracy, efficiency, and practical applicability of short-term wind power forecasting and early wind turbine fault detection using SCADA and meteorological forecast data. The first part of the dissertation develops and investigates a virtual-sensor-based method for condition monitoring and early fault detection in wind turbines using SCADA time-series data, including the selection of the most informative features and the evaluation of factors affecting prediction accuracy. The second part of the dissertation analyzes and optimizes recurrent neural-network structures for the virtual sensor by evaluating feature-sequence formation, training schemes, and alternative activation functions to increase accuracy and reduce the computational cost relevant for practical deployment. The third part of the dissertation develops and investigates a bidirectional long short-term memory (BiLSTM) based method for short-term wind farm power forecasting using meteorological forecast data, and evaluates the impact of different numerical weather prediction (NWP) sources and the suitability of an objective function with a normalized Nord Pool price multiplier for day-ahead energy production forecasts. The dissertation contributes to the fields of wind energy and artificial intelligence by proposing and validating data-driven methods for virtual sensing, residual-based early fault detection, recurrent-model optimization, computationally efficient activation-function selection, and economically meaningful short-term wind power forecasting. The research results have been published in three peer-reviewed scientific journals and one conference proceeding, and were presented at seven conferences and seminars.
Doctoral dissertation readers can search via VILNIUS TECH Virtual Library.
