基于双路神经网络的TBM实时性能预测研究

Research on Real-time Performance Prediction of TBM Based on Dual Neural Networks

为了通过施工中易于获取的数据,实时预测岩石掘进机(Tunnel boring machine,TBM)性能从而为掘进参数的选择提供参考,依托广东某输水隧洞的地勘资料和掘进数据,分析主要掘进参数和场切深指数在不同类型围岩中的分布规律,在围岩类型由Ⅱ类变为Ⅴ类的过程中,刀盘扭矩、刀盘推力、贯入度等参数均存在着均值和标准差同时减小的趋势;除刀盘驱动系统外,支护系统和排渣系统中的参数的均值和标准差与场切深指数存在显著相关性。利用隐马尔可夫模型计算隧洞沿程各类围岩的概率分布,将其作为预测模型特征可提高预测精度但会增加过拟合风险。提出将掘进参数和围岩信息分别输入的双路神经网络模型,模型可对TBM施工前方一环的平均掘进性能进行预测,并将其与经典机器学习模型进行比较。结果表明,所提出的性能预测模型可以保证预测精度优于经典机器学习模型的同时拥有良好的泛化能力。

With the goal of predicting tunnel boring machine(TBM)performance in real time through data that is easy to obtain during construction,relying on the geological survey data and excavation data of a water conveyance tunnel in Guangdong,the distribution of the main boring parameters and field depth index in different types of surrounding rock is analyzed.The hidden Markov model(HMM)is used to calculate the probability distribution of various surrounding rocks along the tunnel,and using it as a prediction model feature can improve the prediction accuracy but increase the risk of overfitting.A two-way neural network model is proposed,which predicts the average advance performance of the first ring in TBM construction and compares it with classical machine learning models.The results show that the proposed performance prediction model can ensure that the prediction accuracy is better than that of the classical machine learning model and has good generalization ability.