Research on Virtual Temperature Measurement Method for High‐Level Radioactive Liquid Waste in Cold Crucible Glass Melts Based on Hybrid Neural Network Model
Science and Technology of Nuclear Installations · 2026-01 · 2026 (1)
DOI: 10.1155/stni/3911456 ↗Kurz gefasst
Fachartikel in der Zeitschrift Science and Technology of Nuclear Installations, online erschienen am 19.08.2026. Verfasst von Zexue Wang und 5 weiteren Autor:innen. Die Arbeit nennt 19 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
In the glass solidification process of high‐level radioactive liquid waste (HLLW) in cold crucibles, the temperature of glass melt is a key parameter that determines the quality of glass matrix and the stable operation of the process. However, under the extreme coupled conditions of high temperature (> 1200°C), strong radiation, and severe corrosion, traditional contact‐type thermocouple temperature measurement devices are confronted with problems such as high‐temperature melting, severe corrosion, and mechanical fracture, resulting in short service life and high frequency of maintenance and replacement, which seriously restricts the stable operation of the process. To overcome the above limitations, this paper proposes a cold crucible virtual temperature measurement model based on a temporal neural network. The model takes the high‐frequency power supply current and voltage, stirrer rotation speed and position, and historical temperature data collected during the 2000‐h continuous operation of the cold crucible engineering prototype as training samples and constructs a temporal neural network with large‐delay characteristics. This network can predict the glass melt temperature at the next moment based on the process data of the previous 5000 s, with a mean squared error (MSE) of 2.54 × 10 −3 and a prediction accuracy of about 85% on the test set. To further improve the prediction performance, this study designs a two‐stage hybrid neural network, which adds a short‐term memory network on the basis of the original temporal neural network. This hybrid model can correct the temperature at the next moment based on the process data of the previous 500 s and the corresponding predicted temperature. It should be noted that during the start‐up phase of this hybrid model, it is necessary to rely on the operation of a physical temperature measurement device for at least 500 s and use the temperature data output by the device as the model input to initiate temperature prediction. Experimental results show that the virtual temperature measurement model achieves an MSE of 8.97 × 10 −4 and a prediction accuracy of about 95% on the test set, which is about 12% higher than that of the ordinary temporal neural network. This method effectively avoids the limitations of physical temperature measuring instruments in extreme environments, provides a new high‐reliability and low‐maintenance–frequency temperature measurement approach for the HLW glass solidification process, and has important theoretical and engineering significance for promoting the intelligence and operational safety of nuclear waste treatment equipment in China.
Zitieren
APA
Wang, Z., Zhang, S., Zhang, K., Qie, D., Bao, L., & Tan, S. (2026). Research on Virtual Temperature Measurement Method for High‐Level Radioactive Liquid Waste in Cold Crucible Glass Melts Based on Hybrid Neural Network Model. Science and Technology of Nuclear Installations, 2026(1). https://doi.org/10.1155/stni/3911456
Vancouver
Wang Z, Zhang S, Zhang K, Qie D, Bao L, Tan S. Research on Virtual Temperature Measurement Method for High‐Level Radioactive Liquid Waste in Cold Crucible Glass Melts Based on Hybrid Neural Network Model. Science and Technology of Nuclear Installations. 2026;2026(1). doi:10.1155/stni/3911456
BibTeX
@article{Wang2026,
author = {Wang, Zexue and Zhang, Shengdong and Zhang, Keqian and Qie, Dongsheng and Bao, Liangjin and Tan, Shengheng},
title = {Research on Virtual Temperature Measurement Method for High‐Level Radioactive Liquid Waste in Cold Crucible Glass Melts Based on Hybrid Neural Network Model},
journal = {Science and Technology of Nuclear Installations},
year = {2026},
volume = {2026},
number = {1},
doi = {10.1155/stni/3911456}
}
EUVIMED