Answers are grounded in this paper and cite their sources.
Purpose To address the limitation that existing digital twin models for power transformers lack real-time data and cannot reflect actual thermal conditions, this study aims to propose a thermal digital twin system using distributed optical fiber temperature measurement, enabling accurate condition monitoring and fault prediction. Design/methodology/approach The approach integrates gray wolf optimization with chaotic opposite-based learning and support vector regression for temperature correction. Multi-scale thermal features and temporal dependencies are extracted via an iterated dilated convolutional neural network, bidirectional long short-term memory networks and sparrow search algorithm, all validated against measured data. Findings The algorithm shows strong agreement with measurements (correlation 0.9413, root mean square error 0.23°C). Fault prediction accuracy reaches 98.72% (area under the curve 0.966). Warning lead times are 35.26 min (light load) and 25.81 min (heavy load), with handling success rates of 99.24% and 97.87%, respectively, outperforming comparative systems. Originality/value This work uniquely combines distributed optical fiber sensing with hybrid optimization-learning algorithms to construct a real-time thermal digital twin. It delivers robust adaptability under varying loads and provides a practical framework for intelligent transformer operation and maintenance.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: