基于人工神经网络的地铁施工项目成本与工期延误预测研究  

Research on Cost and Schedule Delay Prediction for Subway Construction Projects Based on Artificial Neural Networks

作　　者：方基飞 FANG Jifei(China Railway 18th Bureau Group Third Engineering Co.,Ltd.,Zhuozhou,Hebei 072750,China)

机构地区：[1]中铁十八局集团第三工程有限公司,河北涿州072750

出　　处：《施工技术(中英文)》2025年第7期57-63,共7页Construction Technology

基　　金：中国铁建股份有限公司2022年度科技研究开发计划(2022-C1);中铁十八局集团有限公司2022年度科研创新项目(C2022-051)。

摘　　要：地铁施工项目因地质条件复杂及动态参数耦合,常面临成本超支与工期延误风险。本研究基于人工神经网络(ANN)构建了成本与工期预测模型,采用反向传播算法(ANN-BP)优化网络结构与激活函数。通过对济南地铁工程数据的验证,研究结果表明,双曲正切函数(tanh)结合5神经元隐藏层的模型在成本和工期预测中均表现出优异的性能。具体而言,成本预测的测试集决定系数(R~2)为0.899,均方根误差(RMSE)为0.028;而工期预测的R~2达0.971,RMSE为0.024,较Sigmoid函数误差显著降低了53.8%。与支持向量回归(SVR)和随机森林(RF)模型相比,ANN-BP在成本预测中的R~2分别提高了18.7%和9.2%,RMSE降低了32.4%和22.2%;在工期预测中的R~2分别提高了21.5%和14.0%,RMSE降低了28.9%和22.6%。模型通过融合岩土力学参数与盾构动态工况数据,有效捕捉了围岩等级、推力扭矩等非线性关系,误差集中于±5%区间内。本模型为隧道工程成本控制与进度管理提供了高精度工具,其算法鲁棒性在复杂地质场景中具有显著应用价值。Subway construction projects often face the risk of overruns of cost and delays of construction period due to complex geological conditions and dynamic parameter coupling.In this study,a model for cost and construction period prediction was constructed based on an artificial neural network(ANN),and the back propagation(ANN-BP)algorithm was used to optimize the network structure and activation function.Through the verification of engineering data of Jinan Metro,the research results indicate that the model of hyperbolic tangent function,combined with 5 neuron hidden layers,shows excellent performance in cost and construction period prediction.Specifically,the coefficient of determination(R2)of the test set for cost prediction is 0.899,and the root mean square error(RMSE)is 0.028.The test set R2 of the construction period prediction is 0.971,and the RMSE is 0.024,which is significantly reduced by 53.8%compared with the Sigmoid function.Compared with support vector regression(SVR)and random forest(RF)models,R2 of ANN-BP in cost prediction is increased by 18.7%and 9.2%,respectively,and RMSE is decreased by 32.4%and 22.2%,respectively.In the construction period prediction,R2 is increased by 21.5%and 14.0%,respectively,and RMSE is decreased by 28.9%and 22.6%,respectively.The model effectively captures the nonlinear relationship between surrounding rock grade and thrust torque by integrating geotechnical mechanical parameters and shield dynamic working condition data,and the error is concentrated in the range of±5%.This model provides a high-precision tool for cost control and schedule management of tunnel engineering,and its algorithm robustness has significant application value in complex geological scenarios.

关 键 词：地铁 隧道 人工神经网络 成本 工期 双曲正切函数 

分 类 号：TU712.4[建筑科学—建筑技术科学]