TL;DR
This study introduces a shift from traditional classification toward regression‑based prediction of fault severity through a novel integration of Support Vector Regression and Histogram‑Based Gradient Boosting, forming a scalable tool for continuous fault monitoring and strengthens grid resilience by enabling proactive maintenance strategies.
A robust Machine Learning (ML)-based framework for accurately locating electrical faults in wind farm collector networks and achieves the highest accuracy, with prediction errors not exceeding 2%.
The findings show that the suggested hybrid model works better than conventional techniques, with a fault classification accuracy of 98.66% as opposed to decision trees’ 97.42% and SE-CDAE’s 97.98% accuracy.
: With the deepening integration of renewable energy and smart grid technologies, the accurate prediction of power equipment failures and assessment of their health status have become increasingly critical. Current methodologies often operate in isolation: generalized linear models excel in static health assessment but fail to capture temporal dynamics, while time series models forecast trends but lack mechanistic insight into failures. This paper bridges this gap by proposing a comprehensive hybrid model that fuses the interpretability of GLMs with the predictive power of time series analysis. Our approach begins with rigorous data cleaning and preprocessing of historical sensor data. We then implement a two-stage strategy: Stage 1 employs a Poisson regression model to estimate a foundational health index, translating static operational conditions into an interpretable score. In Stage 2, this health score sequence is treated as a new time series and modeled jointly by an ARIMA component for linear trend analysis and an LSTM network for learning non-linear patterns, resulting in a robust fault trend prediction. Experimental results on real operational datasets confirm that our model achieves superior performance compared to all baseline models, showing significant reductions in RMSE and improvements in F1-score and AUC. The framework effectively provides earlier warnings of sub-healthy states, offering a practical and innovative solution for intelligent, data-driven maintenance strategies in the power industry.
This article explores machine learning techniques (MLTs) as a modern alternative to enhance the interpretation of DGA data for early-stage fault detection in service transformers, and demonstrates that random forest and gradient boosting outperform others, achieving up to 98% accuracy.
Nigeria’s electricity system continues to experience reliability challenges arising from transmission and distribution faults, protection coordination difficulties, voltage and frequency deviations, ageing assets, and uneven network observability. Machine learning (ML) is increasingly proposed as a means of identifying abnormal operating states earlier and supporting faster, more selective intervention. This critical narrative review evaluates the extent to which the international fault-analytics literature can support genuine fault prediction and reliability improvement in the Nigerian smart-grid context. Literature published from 1 January 2010 to 22 June 2026 was examined, while older foundational Nigerian evidence was retained where it remained necessary for context. The evidence shows a pronounced terminological and translational gap. Most studies labelled as predictive apply ML after fault inception to detection, classification or localisation, commonly on simulated or synthetically augmented data. Convolutional, recurrent, ensemble, support-vector and hybrid methods often achieve very high test performance in controlled settings, but these results do not establish prospective prediction, field robustness or reduction in outage burden. Nigeria-specific evidence demonstrates technical feasibility for high-impedance-fault diagnosis and voltage-stability classification, yet field-validated prospective models linked to reliability outcomes remain scarce. The strongest near-term opportunity is therefore not autonomous fault anticipation in the abstract, but staged deployment of ML for condition-based risk scoring, abnormal-event recognition, protection support and maintenance prioritisation using utility data streams. Successful translation will require dependable sensing, labelled event histories, topology-aware modelling, uncertainty estimation, cybersecurity, model-drift governance and prospective evaluation against operational endpoints such as outage frequency, interruption duration, energy not supplied and protection misoperations. ML can contribute meaningfully to Nigerian grid reliability, but only when treated as one component of an engineering, data and governance architecture rather than as a substitute for protection design, asset renewal or network reinforcement.
Reliable fault detection in nonlinear process systems requires both accurate classification and interpretable analysis of faults with weak or near-normal signatures. This study develops an explainable, data-driven, and temporally informed multiclass framework for the Tennessee Eastman Process (TEP), retaining all 21 operating conditions (20 fault types and the normal operating condition). Six sliding-window statistics were extracted from 52 process variables and classified using Extreme Gradient Boosting (XGBoost). Performance was evaluated at both sample and run levels through fault-specific window analysis, an a priori validation-driven hierarchical decomposition, SHapley Additive exPlanations (SHAP), a Relative Sensitivity Index (RSI) based on detection-delay sensitivity, and computational benchmarking. Aggregating sample-level predictions (macro F1 = 0.8823) into run-level decisions via majority voting improved performance substantially (macro F1 = 0.9515). Across five seeded repetitions, the mean run-level macro F1-score was 0.9522±0.0035 (95% CI: ±0.0044). Under the primary evaluation, 18 of 21 classes achieved F1 ≥0.97. Fault 3 benefited strongly from extended temporal context, whereas Normal operation, Fault 9, and Fault 15 retained a structured but asymmetric confusion pattern dominated by Normal–Fault 9 errors. SHAP identified model-specific attribution patterns associated mainly with cooling-water-related variability features, while RSI indicated greater prediction-stream sensitivity for the historically difficult faults. Feature extraction and inference required approximately 6.2 ms on CPU and 35.3 ms on GPU, negligible relative to the 180 s sampling interval. These findings indicate that temporal context, run-level aggregation, and explainability can jointly support accurate data-driven fault diagnosis while revealing persistent fault-specific ambiguity.
We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.
