基于BLSTM-AM模型的TBM稳定段掘进参数预测

Predicting boring parameters of TBM stable stage based on BLSTM networks combined with attention mechanism

隧道掘进机(TBM)掘进数据的上升段为实时岩体条件感知和掘进性能参数预测提供了丰富的信息。提出一种融合注意力机制的双向长短时记忆网络,来实现TBM掘进稳定段的性能参数预测。在所提出的模型中,4个主要参数的时间序列数据作为主要输入来提取岩机作用关系,稳定段的推进速度和刀盘转速作为辅助输入来考虑主司机的控制行为,模型输出推力和扭矩预测值。不同于传统的预测模型,所提出的模型不依赖于地质参数,通过自动学习上升段的特征来建立控制参数与预测性能参数之间的映射关系。模型建立过程,采用多项数据处理技术来修正异常值、过滤噪声及归一化等,并提出了基于扭矩时序曲线来识别TBM上升段和稳定段的方法。依托于吉林引松供水隧洞工程,验证了该模型的有效性和准确性。结果表明,所建模型有较好的预测效果,可辅助于类似地质条件的TBM智能化施工。

The ascending branch of tunnel boring machine(TBM)tunnelling data provides rich information for real-time rock mass condition perception and the prediction of boring performance parameter.This paper proposed a bidirectional long short-term memory network combined with attention mechanism to predict the performance parameters in the stable phase of TBM tunneling.In our model,time series data of four main parameters are taken as the main input to extract the rock-machine interaction relationship,and the advance speed and RPM given in the stable phase are taken as auxiliary inputs to consider the human control behavior,and the output of the model is the predicted values of thrust and torque.Different from the traditional prediction model,the proposed model does not require geological parameters,and establishes the mapping relationship between control parameters and performance parameters by automatically learning the characteristics of the ascending branch data.In the process of model establishment,some data preprocessing techniques are used to correct the outlier data,filter noise and normalize data,etc.,and a method for segmenting the ascending and stable branch based on the torque-time curve is proposed.Relying on the Jilin Yinsong water supply tunnel project,the effectiveness and accuracy of the model are verified.The results show that the overall prediction effect of the proposed model is good,which can assist the intelligent construction of TBM with similar geological conditions.