基于WOA-LightGBM的盾构切削锚杆参数预测

Prediction of Shield Performance in Cutting Anchor Bolts Based on WOA-LightGBM

摘要:

在对锚固地层工程特性分析的基础上，提出了6个可指代锚固地层特性的工程相关指标，将3个盾构机可操作参数一并纳入输入特征，并以盾构机贯入度和刀盘扭矩作为盾构机掘进性能的输出指标，构建了一套适用于盾构机穿锚问题预测的模型指标。依托武汉地铁实际工程，收集了盾构机穿锚实时掘进数据，采用LightGBM方法分别搭建了贯入度和刀盘扭矩预测模型，并利用鲸鱼优化算法（WOA）对LightGBM内的超参数进行寻优，最终得到WOA-LightGBM预测模型。结果表明，构建的盾构机穿锚模型指标具有一定的合理性，可成功预测盾构机穿锚掘进性能；与传统BP、ELM神经网络相比，WOA-LightGBM预测模型耗时相近，在预测精度方面有着明显优势。

Abstract:

On the basis of the analysis of engineering characteristics of anchoring strata， six relevant engineering indicators referring to characteristics were proposed. Three operational parameters of shield machine were included into inputs. A set of model indexes for the prediction of shield performance was constructed by taking the penetration rate and the torque of cutter head as outputs of tunneling performance. In combination with an actual project of Wuhan metro，the excavation parameters were collected during the anchor-cutting excavation. Prediction models for penetration rate and torque were developed respectively based on LightGBM method， and whale optimization algorithm （WOA） was used to find the most suitable parameters in LightGBM model. Results show that the proposed model indexes are proper and rational， which can predict the shield performance when tunnelling in anchorage zone； the WOA-LightGBM-based prediction model has obvious advantages in prediction accuracy compared with BP and ELM neural network， while the time-consuming is similar.