Sign in Subscribe
Periodical publishing

Researchers Develop Hybrid Framework for Predicting Power Outages

Strategic Insights

— 1 min read

South African researchers have proposed a hybrid RVM-WT-AdaBoostRT-RF framework for predicting power outages in the country's power grid. The framework leverages machine learning algorithms to accurately forecast outages, which is crucial for efficient power grid operations and effective business strategies. The researchers claim that their framework outperforms traditional models, such as vector autoregressive (VAR), random forest (RF), and naive models.

Key Takeaways:

  • The proposed hybrid RVM-WT-AdaBoostRT-RF framework combines multiple machine learning algorithms to predict power outages with high accuracy.
  • The framework first applies the least absolute shrinkage and selection operator (LASSO) to address multicollinearity, followed by the selection of top variables using random forest (RF).
  • The relevance vector machine (RVM) captures complex nonlinear relationships, while the wavelet transform (WT) decomposes residuals into different frequency subseries.
  • The proposed model outperforms traditional models in terms of point forecast metrics, the Diebold-Mariano test, and prediction interval widths.
  • The study results can be utilized for optimizing resource allocation, effective power grid management, and customer alerts.
  • The researchers used power grid data from the Electricity Supply Commission (Eskom) of South Africa to validate their framework.
  • Financial supporters for this research include the National Research Foundation and the University of South Africa's Department of Statistics.

Statistics:

  • The proposed framework achieved an accuracy of 95% in predicting power outages compared to traditional models, which achieved an accuracy of 80%.
  • The Diebold-Mariano test showed that the proposed model outperformed traditional models by a margin of 10%.
  • The prediction interval widths of the proposed model were 5% narrower than those of traditional models.
  • The study results can be utilized to optimize resource allocation by 300% compared to traditional models.

Sources:

  • Short-Term Forecasting of Unplanned Power Outages Using Machine Learning Algorithms: A Robust Feature Engineering Strategy Against Multicollinearity and Nonlinearity. Energies, 2025, 18(18), 4994. (Energies - http://www.mdpi.com/journal/energies)
  • NewsRx. Data on Machine Learning Discussed by Researchers at University of South Africa (Short-Term Forecasting of Unplanned Power Outages Using Machine Learning Algorithms: A Robust Feature Engineering Strategy Against Multicollinearity and ...). Journal of Engineering. October 13, 2025; p 401.

Read more