SCADA-based thermal degradation prognostics for hydropower step-up transformers: a case study of Nam Theun 2 hydropower plant

Continuous quantification of transformer thermal-aging indicators is vital for risk-informed asset management and maintenance planning in major hydropower facilities, particularly those operating under demanding tropical ambient conditions. However, supervisory control and data acquisition (SCADA)-only applications for transformer prognostics remain underexplored, while conventional approaches often rely on offline chemical diagnostics or invasive sampling, limiting their suitability for continuous real-time monitoring within SCADA environments. This study presents a non-intrusive, data-driven framework for predicting an IEEE Std C57.91-derived equivalent transformer loss of life (LOL) indicator using routinely available operational SCADA parameters, validated on continuous multi-year data from Nam Theun 2 hydropower plant (HPP) in Lao PDR (capacity 1,080 MW). By bridging the IEEE C57.91 thermal-aging formulation with dynamic temporal feature engineering, including differential aging states and autoregressive thermal lags, the proposed methodology captures nonlinear aging dynamics associated with rapid load variations and tropical ambient thermal fluctuations. Twelve generator step-up (GSU) transformers and seven regression paradigms under chronological partitioning strategies, tree-based ensemble models generally provided strong performance. Specifically, the Random Forest model under a 70:30 train–test split ratio achieved a fleet-wide Mean Absolute Error (MAE) of 2.00 and an R 2 of 0.9865 without requiring offline diagnostic data as model inputs. The developed framework provides a cost-effective, scalable basis for continuous SCADA-based thermal-aging monitoring and condition-based maintenance (CBM) across large-scale renewable energy infrastructure.