EUVIMEDEUVIMED entdecken
Crossref

A GMDH Neural Network Approach for Short-Term Load Forecasting in Electrical Substation Using Hourly Time Data

Kenrick Agustin Secugal, Jomar Consorte, Roy Callope

Journal of Innovative Technology Convergence · 30.06.2026 · 8 (2) · 19-28

DOI: 10.69478/jitc2026v8n2a003 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Innovative Technology Convergence, online erschienen am 30.06.2026. Verfasst von Kenrick Agustin Secugal und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate short-term load forecasting (STLF) is essential for optimizing power dispatch and preventing equipment failures in localized electrical grids. This study developed and validated a reliable predictive model for the Antique substation using the Group Method of Data Handling (GMDH) machine learning algorithm. The model was trained and tested on 692 hours of historical electrical load data, split into 592 hours for training and 100 hours for out-of-sample testing. The self-organizing network used a linear transfer function and reached an optimal polynomial complexity of 72 with 48 selected variables. After thorough validation, the algorithm forecasted grid demand for a 100-hour unseen horizon (September 2 to September 6). Performance metrics show exceptional predictions and complete absence of data over fitting. The training phase yielded a Mean Absolute Percentage Error (MAPE) of 3.28%. Accuracy successfully improved during the blind testing phase, resulting in a MAPE of just 2.70%, a Root Mean Square Error (RMSE) of 81.46% kW, and a Mean Absolute Error (MAE) of 66.84 kW. For the future forecasting horizon, the model predicted a peak power demand 4,117 kW while establishing strict statistical confidence band of ±142,35 kW. Falling below the standard threshold for STLF, the GMDG algorithm proved to be a highly dependable tool. Integrating this predictive model into daily substation operations will allow engineers to proactively anticipate peak surges, schedule maintenance safety, preventing catastrophic overloads.

Zitieren

APA

Secugal, K. A., Consorte, J., & Callope, R. (2026). A GMDH Neural Network Approach for Short-Term Load Forecasting in Electrical Substation Using Hourly Time Data. Journal of Innovative Technology Convergence, 8(2), 19-28. https://doi.org/10.69478/jitc2026v8n2a003

Vancouver

Secugal KA, Consorte J, Callope R. A GMDH Neural Network Approach for Short-Term Load Forecasting in Electrical Substation Using Hourly Time Data. Journal of Innovative Technology Convergence. 2026;8(2):19-28. doi:10.69478/jitc2026v8n2a003

BibTeX
@article{Secugal2026,
  author = {Secugal, Kenrick Agustin and Consorte, Jomar and Callope, Roy},
  title = {A GMDH Neural Network Approach for Short-Term Load Forecasting in Electrical Substation Using Hourly Time Data},
  journal = {Journal of Innovative Technology Convergence},
  year = {2026},
  volume = {8},
  number = {2},
  pages = {19--28},
  doi = {10.69478/jitc2026v8n2a003}
}
Kostenlos starten →