Weather Feature Selection for Robust and Optimized Energy Load Prediction | ||||
| Journal of Engineering Advances and Technologies for Sustainable Applications | ||||
| Volume 1, Issue 1, January 2025, Page 1-16 PDF (995.66 K) | ||||
| Document Type: Original research paper | ||||
| DOI: 10.21608/jeatsa.2025.427790 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
Mohsen Tavakolian  ; Hamidreza Zareipour
| ||||
| Department of Electrical and Software Engineering, University of Calgary, Calgary, AB, Canada | ||||
| Abstract | ||||
| This study explores the impact of weather features on short to medium electricity load prediction across diverse geographical locations. Using hourly load data, we evaluated the effectiveness of several feature selection methods, including Mutual Information (MI), Principal Component Analysis (PCA), Lasso, and Heatmap correlation. We benchmarked these feature selection methods with a hybrid deep learning model to investigate the impact of choosing the correct multiple weather features instead of temperature. For this purpose, we practiced different combinations of temperature, relative humidity, dew point, air pressure, and wind speed benchmarked with the base case of single feature (temperature). The comparison was performed based on the load prediction accuracy improvement. The hybrid Artificial Neural Network (ANN) and temporal setup was implemented to predict energy loads across four different lead times (1, 6, 12, and 24 hours ahead) to not only study the feature selection methods, but also its behavior at different lead time predictions. Moreover, this study inspected the dynamic behavior of weather features selection by location to explore the need for location-specific feature engineering. All steps and theories were examined by a real-world dataset from a location of interest and the result was visualized across the geographical extent, offering insights into the spatial variability of feature importance. Future work will investigate the development of lead-time-specific models to further improve load prediction accuracy. This research highlights the importance of an in-depth inspection of weather feature selection and its dynamic behavior for enhancing energy load forecasting models. | ||||
| Keywords | ||||
| Energy load prediction; Feature selection; Weather parameters; Machine learning; Artificial Neural Network (ANN); Temporal models | ||||
|
Supplementary Files
|
||||
|
Statistics Article View: 98 PDF Download: 48 |
||||
