军事知识服务平台
盾构掘进参数跨工况在线增量预测方法
Cross-condition online incremental prediction method for shield tunnelling parameters
摘要 盾构掘进过程中,不同工况间的监测数据分布差异显著,离线预测模型对新工况适应性不足;仅利用少量样本在线更新时,又易出现震荡与过拟合.为此,提出一种掘进参数跨工况在线增量预测方法,从工程含义一致的共享掘进参数中识别稳定候选关系,构建共享预测器;利用新工况本地掘进参数的低秩表征构建线性残差修正项,并以边界保护控制在线修正幅度;引入在线专家加权机制,自适应分配不同历史知识继承方式的权重.以总推力和刀盘扭矩为预测目标,在真实盾构掘进数据上逐环在线预测验证.结果表明,所提方法的平均R2和最差R2分别达到0.647和0.575,优于二阶自回归模型等时序基线与常规增量更新策略;在20%输入扰动下,其平均R2和最差R2仍保持在0.611和0.478.该方法无需回放历史工况的环级原始样本,可支撑掘进过程中的逐环在线更新,为跨工况掘进参数预测提供了一种低数据依赖方式.
Abstract:
During shield tunnelling,monitoring data distributions vary significantly across operating conditions,causing poor adaptability of offline models to new conditions;online updating with limited samples is prone to os-cillation and overfitting.To address these challenges,this paper proposes a cross-condition online incremental pre-diction method for tunnelling parameters.Stable candidate relationships are identified among shared tunnelling pa-rameters with consistent engineering meanings to construct a shared predictor.A linear residual correction term is built based on the low-rank representation of local tunnelling parameters from the new condition,with boundary protection constraining the magnitude of online corrections.An online expert weighting mechanism is introduced to adaptively assign weights to different forms of inherited historical knowledge.Using total thrust and cutterhead torque as prediction targets,rolling one-ring-ahead online prediction verification is conducted on real shield tun-nelling data.The results show that the proposed method achieves an average R2 of 0.647 and a worst-case R2 of 0.575,outperforming temporal baselines including a second-order autoregressive model and conventional incre-mental updating strategies.Under 20%input perturbation,the average and worst-case R2 remain at 0.611 and 0.478.The method requires no replay of ring-level raw samples from historical conditions,supports ring-by-ring online updating during tunnelling,and provides a low-data-dependency approach for cross-condition tunnelling parameter prediction.
刊 名 现代隧道技术 ISTICPKU
年,卷(期): 2026, 63(5)
分类号: U455.43
机标分类号: U455.43 U25 TH117.1
在线出版日期: 2026年11月4日
基金项目: 2024YFC3809400:国家重点研发计划2024BGA001:湖北省中央引导地方科技发展专项,