A hybrid optimized data-driven intelligent model for predicting short-term demand of distribution network
- Publisher:
- ELSEVIER
- Publication Type:
- Journal Article
- Citation:
- Sustainable Energy Technologies and Assessments, 2024, 67
- Issue Date:
- 2024-07-01
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
An advanced deep learning-based framework is presented in this study, utilizing sequential neural architecture to enhance precision in short-term load forecasting of low-voltage distribution networks. A three-stage paradigm for precise forecasting is presented, beginning with a generalizing data preprocessing approach, followed by multivariate feature construction and selection, and finally model hyperparameter modification. The proposed model employs feature engineering and clustering techniques, with the former being used to process historical load data, electricity prices, and ecological variables (temperature, dew point, wind speed, and humidity), and the latter, to extract highly correlated features as final inputs. The model's robustness is ensured by careful exploration and optimization of hyperparameters, and the model after post-optimization achieves a notable Mean Absolute Percentage Error (MAPE) of 0.57%, 0.99%, and 1.2% for 5, 15, and 30 min ahead forecasts, respectively. A detailed comparison with other deep learning algorithms reveals that the suggested model consistently outperforms them in anticipating load demands at different time intervals. This designed approach not only highlights the impact of the presented data-driven model but also conveys useful ideas to strengthen energy management in distribution networks.
Please use this identifier to cite or link to this item: