变分贝叶斯在线更新预测锅炉水冷壁温度

打开文本图片集

中图分类号：TB9 文献标志码：A文章编号：1674-5124（2025）09-0191-10

Abstract： Water wall tube in the boiler operation process， due to the high temperature caused by the burst pipe accident occurred from time to time，therefore， to reduce the occurrence of accidents，water wall tube temperature changes in the rapid and accurate prediction is crucial. Considering the strong temporal correlation for processes， this paper develops predictive models based on deep learning methods such as GRU，Informer， and TCN.To cope with the high volatility of operating conditions，a variational Bayesian last layers （VBLL） online updating mechanism is introduced， in which the parameters of allayers except the output layer are kept fixed，allowing only the output layer parameters to be updated in real time. Folowing integration with the VBLL update mechanism， the accuracy of allmodels improved significantly.The TCN model achieved 100% accuracy in predicting errors within 5∘C and 3∘C respectively under smooth working conditions.the accuracy of the GRU model within the 5∘C and 3∘C error ranges is improved from 95.70% and 77.61% to 100.00% and 99.80% ， respectively. Following the integration of the VBLL update mechanism under fluctuating operating conditions，the accuracy of the GRU and TCN models for 5∘C error increased from 22.44% and 30.18% to 70.96% and 62.55% ，respectively. The proposed method significantly improves the model robustness with good adaptability and computational efficiency while enhancing the prediction accuracy.

Keywords： temperature prediction; variational Bayesian; deep learning; online update

0 引言

能源是人类社会生存发展、现代工业体系稳定运行以及居民生活质量提升的重要支撑和基础保障，电力以其高效、清洁和使用便捷等特点，已成为核心的能源之一[1]。（剩余11970字）