高温热管凭借卓越的传热性能和固有非能动安全特性，已成为先进核反应堆的关键传热元件。准确预测其冷态启动过程对保障反应堆安全运行具有重要意义，然而该过程涉及复杂的多相流动与相变传热，传统数值方法难以兼顾计算效率与预测精度。针对此问题，本研究提出一种基于时序深度学习模型的高温热管冷态启动温度场瞬态预测方法。通过高精度数值模拟构建多工况温度场时空演化数据库，采用线性归一化与滑动窗口法构建时序训练样本，建立嵌入LSTM/GRU时序模块的U-Net预测模型以增强动态特征提取能力，并分析了训练集规模和噪声对预测性能的影响。结果表明:改进的U-Net-LSTM和U-Net-GRU模型瞬态预测最大误差约为5 K，稳态预测最大误差不超过1 K；在训练样本有限或噪声较大时仍保持较高精度。本研究为核反应堆中高温热管的实时监测与安全评估提供了高效可靠的计算解决方案。

High-temperature heat pipes have become key heat transfer components in advanced nuclear reactors due to their excellent heat transfer performance and inherent passive safety characteristics. Accurate prediction of the frozen startup process is of great significance for ensuring the safe operation of nuclear reactors. However, this process involves complex multiphase flow and phase-change heat transfer, making it difficult for traditional numerical methods to balance computational efficiency and prediction accuracy. To address this issue, this study proposes a transient prediction method for the temperature field during cold startup of high-temperature heat pipes based on temporal deep learning models. A spatiotemporal evolution database of temperature fields under multiple operating conditions is constructed through high-fidelity numerical simulations. Time-series training samples are established using linear normalization and sliding window methods. A U-Net prediction model embedded with LSTM/GRU temporal modules is developed to enhance dynamic feature extraction capability, and the effects of training set size and noise on prediction performance are analyzed. The results show that the improved U-Net-LSTM and U-Net-GRU models achieve maximum errors of approximately 5 K in transient prediction and no more than 1 K in steady-state prediction, maintaining high accuracy even with limited training samples or significant noise levels. This study provides an efficient and reliable computational solution for real-time monitoring and safety assessment of high-temperature heat pipes in nuclear reactors.

中核集团青年英才项目