Distribution Transformer Overload Aging Risk Prognosis for Agricultural Feeders

About Dataset

This dataset and accompanying model address a documented, India-specific failure mode: rural agricultural-feeder distribution transformers fail at roughly 12-18% per year (versus under 1% in mature grids), driven mainly by saturated pump loading during restricted DISCOM agricultural-supply windows, phase/load imbalance, and cumulative insulation aging rather than manufacturing defects. The dataset simulates an hourly IoT sensor stream (ambient temperature, top-oil temperature, winding hot-spot temperature, three-phase current, phase-unbalance factor, oil moisture, oil level) across a heterogeneous fleet of 40 transformers (16-100 kVA, agricultural/mixed/urban-residential feeder types, install years 2008-2023) over a 45-day simulated period, generated from the IEEE Std C57.91-2011 loading-guide thermal-aging equations plus a phase-imbalance stress extension. Two labelled ML tasks are provided: a risk-band classification (healthy/watch/warning/critical) and a 24-hour-ahead overload early-warning classification, evaluated with a transformer-grouped train/test split to test genuine generalization to unseen assets. The dataset is synthetic-but-physics-grounded (no real utility data was available at this sensor resolution and labeling depth); it is intended as a research/prototyping asset and as a template pipeline to be retrained on real interval-metered load and weather data before operational deployment.