Risk-Based Dynamic Thermal Rating in Distribution Transformers via Probabilistic Forecasting

A team of researchers led by Scott Angus has developed a novel AI framework that enables distribution network operators to safely extract more capacity from existing low-voltage transformers. The system addresses a critical infrastructure challenge: rising electricity demand coupled with long lead times and high costs for transformer replacements. By moving beyond static, conservative protection devices, their probabilistic forecasting approach predicts day-ahead thermal protection settings that maximize usable capacity while quantifying overheating risk through prediction percentiles.

The core innovation is a clustered quantile regression model trained on historical load, temperature, and metadata from 644 UK low-voltage transformers. This AI model directly predicts the optimal scale factor for dynamic thermal ratings, allowing operators to make risk-informed decisions. Results demonstrate a consistent 10-12% capacity gain compared to traditional static settings, with actual hotspot temperature risk accurately matching the selected probability threshold—even when accounting for realistic temperature forecast errors.

This research, submitted to the 24th Power Systems Computation Conference, represents a practical operational tool for utilities. By enabling adaptive protection settings, network operators can defer costly capital investments in new transformers while managing grid reliability through quantified risk assessment. The framework transforms how utilities utilize existing assets, turning passive protection into active capacity management on operational timescales.