Safety prediction of shield tunnel construction using deep belief network and whale optimization algorithm

Due to ground loss and shallowly buried tunnels, there are serious safety problems in shield tunnel construction. To comprehensively describe the safety of shield tunnel construction, two safety control indices (ground settlement and segment floating) were applied to represent the two main aspects of construction safety (surrounding environment and tunnel structure). Here, a deep-learning method involving a deep belief network (DBN) optimized by a whale optimization algorithm (WOA) called WO-DBN is proposed to predict ground settlement and segment floating. Based on 370,404 engineering data of shield tunnel construction for Guangzhou subway Line 18 in China, the mean absolute errors of the WO-DBN method for the two indices were only 2.255 and 0.954, respectively. The results show that the WO-DBN achieves a high prediction accuracy, and that it can be effectively used for safety prediction of real shield tunnel construction. The improvement of the WO-DBN, such as through using the newly developed activation functions, should be a future research direction.