Empirical Modeling of Cryogenic System for Hybrid SFCL Using Support Vector Regression

• Published in: Journal of superconductivity and novel magnetism Vol. 26; no. 4; pp. 1265 - 1273
• Main Authors: Seo, In-Yong, Hyun, Ok-Bae, Kim, Heesun, Ha, Bok-Nam, Song, Il-Keun, Chae, Wookyu, Kim, Min-Jee, Kim, Seong-Jun
• Format: Journal Article Conference Proceeding
• Language: English
• Published: Boston Springer US 01.04.2013; Springer
• Subjects: Applied sciences; Characterization and Evaluation of Materials; Condensed Matter Physics; Cryogenic cooling; Electronics; Exact sciences and technology; Faults; Hybrid systems; Magnetic Materials; Magnetism; Mathematical analysis; Original Paper; Physics; Physics and Astronomy; Regression; Semiconductor electronics. Microelectronics. Optoelectronics. Solid state devices; Strongly Correlated Systems; Superconducting devices; Superconductivity; Vectors (mathematics); Support vector regression; Response surface method; Particle swarm optimization; Superconducting fault current limiter; Cryogenic cooling; Prototype; Cooling; Response surface methodology; Neural network; Empirical model; Modeling; Optimization; Fault currents; Current limiter; Cooling system; Copper
• ISSN: 1557-1939; 1557-1947
• DOI: 10.1007/s10948-012-1965-7

The hybrid superconducting fault current limiter (SFCL) is now at the stage of practical use in a power grid in Korea. A cryogenic cooling system was designed, fabricated, and successfully tested for a prototype of 22.9 kV/630 A SFCL. The operation scheme of cryogenic system has been investigated in preparation for temporary loss of cryocooler power in hybrid SFCL (in Kim et al., IEEE Trans. Appl. Supercond. 21(3):1284–1287, 2011 ). In this paper, we investigated the empirical modeling of cryogenic cooling system for SFCL using principal components and auto-associative support vector regression (PCSVR) for the prediction and fault detection of the cryogenic cooling system. For empirical model, data were acquired during a blackout test of cryogenic cooling system. Blackout times of the test were 1 hour and 4 hours at two operation current levels. Three set of data were used for training and optimization of the model and the rest set of data was used for verification. Signals for the model are temperatures measured at copper band and cold head of cryocooler, system pressure and liquid temperatures measured at two locations in liquid-nitrogen pool. For optimization of the SVR parameters, the response surface method (RSM) and particle swarm optimization (PSO) were adopted in this paper. After developing the empirical model we analyzed the accuracy of the model. Also, these results were compared with that of auto-associative neural networks (AANN). RSM and PSO gave almost the same optimum point. PCSVR showed much better performance than AANN in accuracy aspects. Moreover, this model can be used for the prognosis of cryogenic cooling system for SFCL.