关键词: 隧道掘进机（TBM）, 掘进速率, 区间预测, 双向长短时记忆网络, 哈里斯鹰优化算法, Bootstrap方法

Abstract: Previous prediction models of the Tunnel Boring Machine (TBM) advance rate mostly adopted the point prediction method and lacked consideration of the uncertainties caused by the subjective selection of model structure, random parameter setting, and random data noise. This paper develops an interval prediction model of the TBM boring rate based on the Bootstrap method and the improved Harris Eagle optimized bi-directional long short-term memory network (BiLSTM). First, we construct a prediction model based on the Improved Harris Hawks Optimization (IHHO) optimized BiLSTM network, and reveal the correlation and time dependency of the boring rate for the stable section operation on the thrust, torque, speed and other boring parameters of the cutterhead for the rising section operation. This model uses the Harris Eagle algorithm based on chaotic mapping, parameter nonlinearization and chaos search strategy to optimize the hyper-parameters of its BiLSTM network for better modeling efficiency and accuracy. Then, the Bootstrap method is used to quantify its model uncertainty and random uncertainty and to obtain clear and reliable prediction intervals. It has been applied to the Qinling Mountain tunnel project under the conditions of surrounding rock class I-III. The results are compared with those of the BILSTM-HHO model, BiLSTM model and BP neural network model, proving the superiority of our new model.

Key words: Tunnel Boring Machine, excavation rate, interval prediction, bidirectional long short-term memory network, Harris Eagle optimization algorithm, Bootstrap method