Accurate forecasting of critical parameters in power grids, such as electricity load, price, and renewable energy production, plays a pivotal role in implementing demand response strategies and maintaining grid stability within smart grid systems. In this paper, we propose a novel deep learning approach based on an encoder-decoder architecture with an attention mechanism to simultaneously predict these essential parameters. To enhance the model\\\'s ability to capture complex patterns and long-term dependencies in time series data, we incorporate a preprocessing step using the Fast Fourier Transform (FFT), which transforms the data into the frequency domain. This transformation aids in reducing noise and extracting latent patterns, thereby improving the model\\\'s predictive performance. The proposed model is evaluated on real-world electricity consumption datasets from multiple regions, including Austria, Italy, Sweden, and the UK. It is compared against several state-of-the-art time series forecasting models, such as LSTM, CNN-LSTM, LSTM-Attention, and a basic encoder-decoder structure. The results demonstrate that our approach achieves significantly higher accuracy in most cases, particularly for short- to medium-term predictions (12 to 48 hours ahead). Additionally, visualizations of the predictions reveal that the proposed method closely follows the actual trends, confirming its effectiveness in capturing both short-term fluctuations and long-term dependencies. This study contributes to the field by introducing a flexible and robust framework that integrates attention mechanisms and frequency-domain preprocessing to improve time series forecasting in the electricity domain. The proposed approach not only enhances prediction accuracy but also provides valuable insights into the underlying patterns of electricity consumption and production, making it a promising tool for smart grid applications. The source code and implementation details of the proposed model are publicly available on GitHub .

Keywords