Three-Parameter Steady-State Temperature-Rise Prediction for a 12 kV Ring Main Unit Using POD and Gaussian Process Regression
Repeated three-dimensional multiphysics solutions make temperature-rise analysis across ring main unit operating conditions computationally expensive. This study combines volume-weighted proper orthogonal decomposition (POD) with Gaussian process regression (GPR) to predict steady-state temperature-rise fields from current and two independent contact-resistivity multipliers. The database contains 187 simulated conditions. POD rank and coefficient mapping are selected using validation maximum-temperature-rise error, with multilayer perceptron and radial basis function methods as comparators. The selected 32-mode POD–GPR model achieves mean relative errors of 0.0039% for maximum temperature rise and 0.0100% for the full field on the in-domain test set. At 1250 A peak current, the mean absolute relative GPR-to-simulation error across ten temperature-rise groups is 0.00697%. The corresponding simulation-to-experiment and GPR-to-experiment errors are 2.7600% and 2.7595%. The median full-field prediction and reconstruction time is 0.016033 s. The model enables rapid steady-state temperature-rise calculation within the investigated parameter domain.
Authors
- Jiaqi Yu
- Gaoyi Ouyang (ORCID: https://orcid.org/0009-0002-1158-7049)
- Junhui Wu
- Xumei Lan
- Lan Jiang
- Wanting Wang
- Yongqin Lan
Institutions
- Xi'an Jiaotong University (CN)
Publication Details
- Journal
- Electronics
- Published
- 2026-11-16
- DOI
- https://doi.org/10.3390/electronics15225319
- Primary Topic
- Model Reduction and Neural Networks
- Type
- article
- Field-Weighted Citation Impact
- 0.00