青藏高原冻土区的土壤水分动态对区域水文过程和生态系统的稳定性具有重要影响。然而，由于冻土区复杂的气候和地形条件，传统的土壤水分监测和预测方法耗时耗力、实施困难，因此利用深度学习方法模拟高原冻土区土壤含水量成为一种有效手段。基于长短期记忆（long short-term memory, LSTM）模型，通过引入多源数据，包括土壤温度、辐射、气温、降水、相对湿度等关键变量，构建并训练LSTM模型，对青藏高原冻土区的土壤水分进行模拟。研究结果表明，LSTM模型能够有效捕捉冻土区土壤水分的动态变化，较好地模拟不同气象条件下土壤含水量随时间的变化特征；该模型在唐古拉、甜水海、西大滩地区的土壤含水量均方根误差分别为0.018、0.023、0.033，决定系数分别为0.937、0.873、0.986。可见LSTM在处理非线性、时序数据方面具有显著优势，特别是在复杂的冻土环境中体现出较低的误差、较为稳定的数值表现，研究结果为基于深度学习的土壤水分模拟提供了新思路，为揭示冻融条件下土壤含水量变化规律提供参考。

The dynamic of soil moisture in the permafrost region of the Qinghai-Tibet Plateau has a significant impact on the regional hydrological process and the stability of the ecosystem. However, due to the complex climatic and topographic conditions in the permafrost region, traditional soil moisture monitoring and prediction methods are time-consuming, labor-intensive and difficult to implement. Therefore, using deep learning methods to simulate soil moisture content in the permafrost region of the plateau has become an effective approach. Based on the long short-term memory (LSTM) model, the LSTM model was constructed and trained by introducing multi-source data, including key variables such as soil temperature, radiation, air temperature, precipitation, and relative humidity, soil moisture in the permafrost region of the Qinghai-Tibet Plateau was simulate. The results show that the LSTM model can effectively capture the dynamic changes of soil moisture in the permafrost region and simulate the variation characteristics of soil moisture content over time under different meteorological conditions. The root mean square error (RMSE) of the model in Tanggula, Tianshuihai and Xidatang areas are 0.018, 0.023 and 0.033 respectively, and the coefficient of determination (R²) are 0.937, 0.873 and 0.986 respectively. It is concluded that the LSTM model has significant advantages in processing nonlinear and time series data, especially in the complex permafrost environment, it shows lower errors and more stable numerical performance. The result provides a new idea for soil moisture simulation based on deep learning and offers a reference for revealing the variation law of soil moisture content under freeze-thaw conditions.