【摘要】 熔盐空气换热器（Molten Salt-to-Air Heat Exchanger,MSAHX）在熔盐堆系统中起着至关重要的作用，准确预测其状态参数的趋势，对于设备的运行与维护、系统稳定和安全运行具有重要意义。为了快速、高效、准确地预测熔盐空气换热器的关键状态参数，提出了一种基于长短时记忆神经网络（Long Short-Term Memory,LSTM）的预测模型。选择钍基熔盐核能系统（Thorium Molten Salt Reactor Nuclear Energy System,TMSR）综合仿真试验平台的MSAHX为研究对象，并使用树状贝叶斯优化算法（Tree-structured Parzen Estimator,TPE）对预测模型进行了超参数优化。数值结果表明，经过优化的预测模型能够在计算时间极短情况下，迅速而准确地预测MSAHX关键参数的趋势。在测试集上熔盐流量、熔盐出入口温度和内部换热管温度的最大预测相对误差均不超过1%，空气出口温度预测相对误差主要分布在1%以内，最大不超过2%。因此，提出的预测模型拥有快速准确预测MSAHX关键参数的能力，后续将进一步部署和应用于实验设施，同时也为熔盐堆其他设备的关键参数预测提供了参考。更多还原

【Abstract】 [Background] The Molten Salt-to-Air Heat Exchanger(MSAHX) plays a key role in molten salt reactor system. Prediction for key state parameters of MSAHX provides significant assistance in early fault identification and diagnosis, effectively enhancing the stability and safety of the Molten Salt Reactor(MSR) system. [Purpose] This study aims to rapidly, efficiently and accurately predict key state parameters of the MSAHX by utilizing a prediction model based on Long Short-Term Memory(LSTM). [Method] The MSAHX of the Thorium Molten Salt Reactor Nuclear Energy System(TMSR) comprehensive simulation test platform was chosen as the research object. The historical experimental data obtained from long-term operation of the MSAHX at various operating conditions was used as the dataset for the prediction model. Then, Tree-structured Parzen Estimator(TPE) was employed to perform hyperparameter optimization on the prediction model, resulting in the model with best performance on the validation set. Finally, the prediction accuracy and efficiency of the prediction model on the dataset were evaluated using average absolute error, maximum relative error, and prediction time as indicators. [Results] The numerical results indicate that the optimized prediction model can rapidly and accurately forecast key parameters of the MSAHX 180 seconds in advance within mere milliseconds of computation. The maximum relative error of prediction on the test dataset for the internal heat exchange tube temperature, the molten salt flow rate, the outlet and the inlet temperature of the molten salt do not exceed 1%, while for the outlet temperature of air the maximum relative error is below 2%. [Conclusion] The prediction model proposed in this study shows speedy and precise predictive capabilities for key parameters of the MSAHX. Furthermore, the prediction model will be deployed and applied to the experiment of the MSAHX, which will provide valuable reference for the prediction for key parameters of other equipment in molten salt reactor system.更多还原