Abstract:  When earth pressure balance shields tunneling in waterrich coarsegrained grounds, muck spewing often occurs due to the high permeability of the muck. As a result, a model for predicting the permeability coefficient of shield muck is proposed based on the random forest algorithm. First, conditioning parameters, such as the water content, foam injection ratio, bentonite slurry injection ratio, the effective particle size of soil, and the hydraulic gradient are collected as input paraters of the model. The results reveal that while the predicted and the measured values are in the same order of magnitude, the root mean square error is 2.4×10-9 cm/s, and the fitting determination coefficient reaches 0.981 9, indicating good prediction accuracy of the model. Then, the model was applied to the Longquan shield tunnel project in Central Yunnan, China, to determine the risk of water spewing during the shield undercrossing the Panlong river. Consequently, recommended conditioning parameters were provided, which accounted for the recorded achievements like the relativelylow permeability coefficient of improved muck, stable chamber pressure, small influence on the upper bridge structure, and safe/efficient shield tunneling.

Key words: earth pressure balance shield, soil conditioning; permeability coefficient, random forest, prediction model