Deep learning-based short-term electric load forecasting
The power industry is a vital indicator of national economic growth, and in today’s modern society, daily life is inextricably linked to the power system. As such, accurate grid forecasting has become essential for informed decision-making and reliable power system operation. Short-term load forecas...
| Autor: | |
|---|---|
| Formato: | tesis de maestría |
| Fecha de publicación: | 2025 |
| País: | España |
| Recursos: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/441434 |
| Acesso em linha: | https://hdl.handle.net/2117/441434 |
| Access Level: | acceso abierto |
| Palavra-chave: | Renewable energy sources Predictive control Deep learning Neural networks (Computer science) predicció a curt termini, aprenentatge profund, xarxa neuronal convolucional (CNN), xarxa de memòria a curt i llarg termini (LSTM), augment de gradient extrem (XGBoost), transformadors, short-term prediction, deep learning, Convolutional Neural Network(CNN), Long Short-Term Memory Network(LSTM), Extreme Gradient Boosting(XGBoost), Transformers Energies renovables Control predictiu Aprenentatge profund Xarxes neuronals (Informàtica) Àrees temàtiques de la UPC::Energies |
| Resumo: | The power industry is a vital indicator of national economic growth, and in today’s modern society, daily life is inextricably linked to the power system. As such, accurate grid forecasting has become essential for informed decision-making and reliable power system operation. Short-term load forecasting (STLF) plays a key role in power system management, facilitating efficient generation scheduling and economic dispatch. However, due to the nonlinear and temporally complex nature of load data, achieving high forecasting accuracy remains a major challenge. Recent advances in artificial intelligence, particularly deep learning, have significantly enhanced the capabilities of load forecasting. Given the diversity, variability, and structured regularities of electric load patterns—which are influenced by meteorological, social, and economic factors — STLF is inherently a complex nonlinear problem that requires large-scale data processing. Deep learning models are particularly well-suited for such tasks. In this study, we develop and compare five forecasting models: Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), Extreme Gradient Boosting (XGBoost), Transformer, and a hybrid CNN-LSTM model. We use hourly electricity demand data from New York City, along with corresponding weather variables, to train and evaluate the models. To address the limitations of traditional approaches, we propose a deep learning-based framework for short-term load forecasting, and conduct systematic experiments to determine the optimal model architecture. Experimental results show that the hybrid CNN-LSTM model achieves the highest forecasting accuracy, with a Mean Absolute Percentage Error (MAPE) of 0.06%, Mean Absolute Error (MAE) of 0.0056, Root Mean Square Error (RMSE) of 0.0071, and a coefficient of determination (R² ) of 0.9894. These findings highlight the effectiveness of integrating spatial and temporal learning in short-term power load forecasting. |
|---|