Electric Load Forecasting of Residential Building

Abstract

As global energy consumption continues to rise relevant energy sectors and communities must accurately forecast future electricity needs. This foresight is critical for effective planning, preserving the stability of the electricity grid, and avoiding blackouts. Accurate forecasting methodologies are critical in guiding decision-making processes correlated to resource allocation, infrastructure development, and policy formulation.

In this thesis, we used a combination of traditional methods like Linear Regression and advanced techniques such as Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Transformer Machine learning models, as well as Empirical Mode Decomposition (EMD) signal processing. Notably, the use of the Transformer method, a relatively new approach to time series forecasting, delivered particularly promising results. We observed a significant improvement in prediction accuracy after incorporating EMD analysis along with training models.

We used a variety of metrics to evaluate model performance and assess its effectiveness. The metrics used were Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Normalized Root Mean Square Error (NRMSE), and coefficient of determination (R2). Using this diverse set of metrics, we aimed to gain a comprehensive understanding of each model's predictive capabilities.

This comprehensive methodology enabled us to evaluate strengths and shortcomings in several forecasting approaches and make informed decisions about their feasibility for practical application. We wanted to ensure the reliability and robustness of our forecasting models by conducting thorough evaluation utilizing several criteria, allowing for more accurate and informed decision-making in energy management strategies in residential buildings and beyond.