【摘要】 顶管顶进力的可靠预测对保证顶管工程顺利施工十分关键。受地质参数和施工条件等多因素影响，顶进力的发展具有复杂性、随机性和不确定性，进而造成准确预测顶管顶力十分困难。引入门控循环单元(GRU)算法来预测顶进力，首先详细描述GRU预测模型的构建方法，其次对模型中神经元个数和学习速率2项超参数进行探究，以提高模型预测的准确度，最后依托某下穿运河矩形顶管实例，采用GRU模型预测顶进力，并与传统时序预测模型作对比，分析模型预测值与实测值的差异。顶进力数据分析结果表明，GRU预测模型拥有较高的准确度和鲁棒性，对长距离矩形顶管顶进力预测方法的改进和优化具有参考作用。更多还原

【Abstract】 Reliable prediction of pipe jacking force was crucial for ensuring the smooth construction of pipe jacking engineering. Influenced by multiple factors such as geological parameters and construction conditions, the development of jacking force was complex, random, and uncertain; hence, making an accurate prediction of pipe jacking force was very challenging. This study introduced the gated recurrent unit(GRU) algorithm for predicting the jacking force. First, the construction method of the GRU prediction model was described in detail. Then, the study explored two key hyperparameters, the number of neurons and the learning rate, to enhance the model’s prediction accuracy. Finally, relying on a case of a rectangular utility tunnel jacked beneath a canal, the GRU model was applied to predict the jacking force and compared with traditional time series prediction models. The comparison between model predicted and actual measured values was analyzed. The analysis results of jacking force data demonstrated that the GRU prediction model offered high accuracy and robustness, providing a valuable reference for improving and optimizing long-distance rectangular pipe jacking force prediction methods.更多还原