在天然气储气站点，异常井筒温度将严重影响储气作业过程，准确地预测井筒温度对于提升作业过程的安全性具有重要意义。在作业过程关联参数分析的基础上，提出了改进时空图卷积神经网络（advanced spatiotemporal graph convolutional neural network, A-SGCN）的储气库井筒温度预测方法。A-SGCN分别通过GCN和长短期记忆网络（long short term memory network, LSTM）进行空间依赖性建模和时间依赖性建模，结合自适应残差注意力机制有效捕捉时空数据之间的复杂关系，最终实现对温度的准确预测。方法的有效性通过黄草峡储气库2号注采站进行了验证，通过草储1井与草储6井的关联监测参数实现了对草储1井井口温度的准确预测。

Gas storage operations can be significantly impacted by abnormal wellbore temperatures at natural gas storage sites. Accurately predicting wellbore temperatures is of paramount importance for enhancing the safety and efficiency of these operations. Based on the analysis of operational parameter correlations, a gas storage wellbore temperature prediction method was proposed using advanced spatiotemporal graph convolutional neural network (A-SGCN). Both GCN and long short-term memory (LSTM) networks were employed by A-SGCN to capture spatial and temporal dependencies, respectively. Based on this framework, an adaptive residual attention mechanism was incorporated to effectively capture the intricate relationships between spatiotemporal data, ultimately enabling accurate temperature prediction. The effectiveness of the method is validated through its application at the Huangcaoxia gas storage No. 2 injection-production station. Accurate prediction of wellhead temperature at Well No. 1 is achieved through the association of monitoring parameters between Well No. 1 and Well No. 6.