A modified back analysis method for deep excavation with multi-objective optimization procedure

Abstract

Real-time prediction of excavation-induced displacement of retaining pile during the deep excavation process is crucial for construction safety. This paper proposes a modified back analysis method with multi-objective optimization procedure, which enables a real-time prediction of horizontal displacement of retaining pile during construction. As opposed to the traditional stage-by-stage back analysis, time series monitoring data till the current excavation stage are utilized to form a multi-objective function. Then, the multi-objective particle swarm optimization (MOPSO) algorithm is applied for parameter identification. The optimized model parameters are immediately adopted to predict the excavation-induced pile deformation in the continuous construction stages. To achieve efficient parameter optimization and real-time prediction of system behavior, the back propagation neural network (BPNN) is established to substitute the finite element model, which is further implemented together with MOPSO for automatic operation. The proposed approach is applied in the Taihu tunnel excavation project, where the effectiveness of the method is demonstrated via the comparisons with the site monitoring data. The method is reliable with a prediction accuracy of more than 90%. Moreover, different optimization algorithms, including non-dominated sorting genetic algorithm (NSGA-II), Pareto Envelope-based Selection Algorithm II (PESA-II) and MOPSO, are compared, and their influences on the prediction accuracy at different excavation stages are studied. The results show that MOPSO has the best performance for high dimensional optimization task.