Pattern Recognition Algorithms to Enhance the Performance of Energy Management Systems and the Reliability of Power Supply

This thesis investigates the application of pattern recognition algorithms in two fields. The first application focuses on enhancing the performance of energy management systems through advanced forecasting. To this end, a recurrent neural network is implemented to predict renewable energy production. The outcomes of the algorithm are input into an effective energy management system enabling a reliable estimate of dispatching programs for production units of plants and microgrids to minimize the operation costs. Additionally, a second algorithm based on boosting techniques is developed to evaluate the profitability of participating in ancillary service markets. Integrating this model into energy management systems allows the computation of a solution adapted to real-time network contingencies. Besides, long-term analyses are enabled, such as assessing the potential benefits of storage systems to increase the flexibility of microgrids. The second application deals with the development of pattern recognition algorithms to support the nowcasting of extreme weather events, leveraging the link between lightning activity and these occurrences. Two models are designed to detect anomalous lightning activity, with the goal of output support information aiding early warnings. Initially, a Gaussian Process regression model is created to nowcast one hour in advance the lightning activity. Subsequently, to account for the temporal and spatial dynamic nature of meteorological structures, a recurrent neural network with convolutional operators is developed to provide a more adaptive solution. These contributions underscore the effectiveness of pattern recognition in addressing diverse challenges, offering effective tools to improve the integration of renewables in power systems and protect infrastructures for a reliable energy supply by enhancing predictions of extreme weather events.