基于机器学习的地铁盾构隧道长期沉降预测

采用机器学习方法，建立一个基于先期区域地质信息及隧道沉降学习资料的盾构隧道长期沉降预测模型， 并以南京地铁 2 号线盾构隧道为例进行算例分析。结果表明：该沉降预测模型能够筛选主要影响因素，并且能够 寻找最佳监督学习算法和最优参数；在不同监督学习算法中，核支持向量机算法与人工神经网络算法都能使模型 达到较高的精度，然而对其参数的依赖性很高，需要细致的调参才能提高预测精度；以人工神经网络算法作为监 督学习算法，经调参后，沉降预测模型的最终预测准确度可达 0.86，10 倍交叉验证平均准确度为 0.82。

Machine Learning Based Prodiction Model for Long-term Settlement in a Metro-shield Tunnel

Based on advanced regional geological information and tunnel settlement data, a prediction model for long-term settlement in a shield tunnel using a machine learning method was proposed. The model was demonstrated through a case study of the Nanjing Metro Line 2 shield tunnel. The results show that the prediction model can analyze and obtain the main influencing factors of long-term tunnel settlement and determine the best supervised learning algorithm and optimal parameters. Among the different supervised learning algorithms, both the kernel support vector machine (KSVM) algorithm and the artificial neural network (ANN) algorithm can achieve high accuracy but require careful parameter adjustment to improve the prediction accuracy. Using the ANN algorithm as a supervised learning algorithm, and after parameter adjustment, the final prediction accuracy of the model reached 0.86, and the average accuracy of the tenfold cross-validation was 0.82.