World Journal of Engineering Research and Technology

Nsikak E. Udoh*, Nseobong I. Okpura, Kingsley M. Udofia

ABSTRACT

This paper quantitatively assesses the dynamic resilience and reliability of a weak 33 kV distribution feeder using a machine-learning framework applied to 3 years of high-resolution operational data. A Bagged Trees ensemble is employed to classify system states and fault types. At the same time, novel metrics—the dynamic Rolling Resilience Index (RNI), the Reliability Function R(t), and Mean Time Between Failures (MTBF)—are introduced to capture the feeder’s time varying health. The results reveal a steep exponential decay of R(t) to near zero within 200 hours, indicating a vanishing probability of fault-free operation beyond one week. The RNI behaves as a brittle switching signal, oscillating between 0 and 1, reflecting a network with no redundancy and no resilience reserve. A right-skewedlognormal fault-duration distribution shows that although many outages are transient, a long tail of extreme durations (exceeding 300 hours) points to severe logistical failures in repair and restoration. Monthly fault peaks correlate with seasonal stressors (e.g., the rainy season), and frequency deviations spanning 20–65 Hz reveal a profound generation-demand mismatch and a lack of grid inertia. The engineering implication is clear: reactive maintenance is failing, and a shift to proactive, condition based infrastructural hardening is not merely beneficial but necessary to arrest the chronic fragility quantified in this study.

[Full Text Article] [Download Certificate] https://doi.org/10.5281/zenodo.21159604