e_Buah Biblioteca Digital Universidad de Alcalá
Publicación
Contribución de congreso (ConferenceOutput)
An encoder-decoder LSTM network for load forecasting in building power systems
2025 IEEE International Instrumentation and Measurement Technology Conference (I2MTC), 19-22 May 2025, Chemnitz, Germany., Intelligent energy management systems have become a trending topic in the last decade, due to those challenges associated to climate change or energy shortages, among others. Efficient energy management and energy demand response are often related to the availability of accurate prediction models that may provide the expected energy consumption for a household or building early enough. For that purpose, deep learning and deep neural networks are already a suitable approach to deal with large datasets from a building energy consumption, thus making it feasible the definition and training of data-driven models. In this context, this work describes a novel encoder-decoder network based on LSTM (Long Short-Term Memory) layers that is capable of forecasting the weekly energy consumption of a public building, starting from the actual energy values from the previous week as well as from the information about the work days in that week. The proposal has been successfully validated in the School of Engineering from the National University of Mar Del Plata (Argentina), achieving a mean absolute error of 2.68 kW in the prediction of the active power of the building during a test period of eleven weeks., Agencia Estatal de Investigación