Combined Deviation Correction Control Strategy for Full-Face Shaft-Boring Machines Based on an LSTM Model

To address delayed attitude correction, limited adaptability of single-actuator systems, and reduced tunneling efficiency in full-face shaft-boring machines (SBMs), this study proposes a PSO-LSTM-based hybrid steel strand–support shoe attitude correction strategy. A coupled dynamic model with a 45° offset configuration is developed to enable coordinated multi-actuator control. A PSO-optimized Long Short-Term Memory (PSO-LSTM) network is employed to predict inclination deviation over a 5 s horizon, providing anticipatory information for proactive control. Based on this prediction, a hierarchical control strategy with adaptive torque allocation is designed to seamlessly coordinate fine correction via steel strand cables and high-torque correction via support shoes. Simulation results demonstrate that the proposed model achieves a prediction accuracy within ±0.02°. Under inclination conditions of 0.05°, 0.3°, and 1.0°, rapid attitude correction is achieved. Compared with independent support shoe control, the maximum horizontal displacement is reduced from 64 mm, 131 mm, and 160 mm to 6.3 mm, 65 mm, and 100 mm, corresponding to reductions of 90.2%, 50.4%, and 37.5%, respectively. The results further indicate that small-angle deviations can be compensated by the steel-strand system without additional support-shoe operations, while medium- and large-angle deviations can be regulated through coordinated actuation of multiple correction systems according to deviation magnitude. Simulation results demonstrate that the proposed method improves attitude correction performance and dynamic response under the investigated simulation conditions. The proposed framework provides a potential solution for intelligent attitude control of SBMs, while further field validation is required before practical engineering deployment.