Machine Learning-Based Fault Prediction and Smart Grid Reliability: A Study of the Benin Electricity Distribution Company (BEDC) Network in Delta State, Nigeria
Frequent, poorly-predicted faults remain the single largest drag on distribution reliability across much of Nigeria, and Delta State is no exception. This study asks a fairly practical question: can machine-learning models, trained on the fault and operational data that the Benin Electricity Distribution Company (BEDC) already collects, predict feeder faults accurately enough to be useful for planning and maintenance? Drawing on a six-year record (2019–2024) of 4,477 fault incidents across selected 11 kV and 33 kV feeders in the BEDC Delta State network, together with matched weather, load and asset-condition data, I trained and compared four classifiers, namely Random Forest, XGBoost, Support Vector Machine (SVM) and an Artificial Neural Network (ANN). XGBoost performed best, returning an accuracy of 95.8%, an F1-score of 94.8% and an ROC-AUC of 0.972. Ambient temperature, feeder loading and rainfall intensity emerged as the strongest predictors. When the predictions were used to guide a pilot condition-based maintenance schedule, the modelled System Average Interruption Duration Index (SAIDI) fell by roughly 33% and the System Average Interruption Frequency Index (SAIFI) by about 33% relative to the historical baseline. The findings suggest that data the utility already owns can be turned into a workable early-warning tool, and I argue that the main barrier is organisational rather than technical
Authors
- Atumah Progress Ejiro
- Christian Okpako Akperegin
Institutions
- Delta State Polytechnic Otefe-Oghara (NG)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.5281/zenodo.24399330
- Primary Topic
- Power System Reliability and Maintenance
- Type
- article
- Field-Weighted Citation Impact
- 0.00