In highway reconstruction and expansion projects, the impact of the abandoned dreg body produced by the original highway on the new bridge project cannot be ignored. If the abandoned dregs are not handled properly, they will cause great threat to the operation of the new highway as well as to the safety of the surrounding environment and personnel. In this paper, based on a South China highway reconstruction and expansion project, a model of the abandoned dreg body, bridge pile and anti-slide pile is established with the help of finite difference software FLAC3D 5.0. The calculation results show that compared with burying the anti-slide pile, slope cutting is more effective for improving the stability of the abandoned dreg body. Under the action of vertical force intensity, the displacement of the top of the pile increases continuously, and the maximum bending moment of the pile body increases firstly and then decreases. The pile displacement and bending moment of the bridge pile increase significantly in the storm condition compared with the normal condition. In addition, the abandoned dreg body hazard prevention and control technical measures are studied. The impacts of different locations of anti-slide pile burial on bridge piles are analyzed and compared, and it is determined that the optimal location of anti-slide pile burial is the middle of the secondary slope.
In recent years, it has become inevitable to dig underneath existing highways when excavating tunnels. The soil settlement induced by ground excavation may adversely affect existing highways. In this study, a settlement monitoring system is used to obtain the settlement sequence of multiple measurement points on the pavement. Multifractal detrended fluctuation analysis (MF-DFA) is used to focus on analyzing the multiple fractal features of the pavement settlement rate. The results show that the settlement rates of the highway caused by the tunnel excavation and construction process all show multiple fractal characteristics. The fluctuations in the measurement points above and near the entrance of the tunnel are more complex and intense. Based on the moving-average method (MA), convolutional neural network (CNN), and Extreme Learning Machine (ELM), MA-CNN and MA-ELM prediction models are constructed to predict the settlement value sequences of the fluctuating points. The results indicate that the MA-ELM prediction model demonstrates superior predictive performance (with R2 values of 0.956, 0.950, and 0.979 on the test set). Further, with the help of the Dung Beetle Optimizer (DBO), a meta-heuristic algorithm for parameter optimization, the hybrid model DBO-MA-ELM greatly improves the prediction performance (R2 of 0.975, 0.997, 0.998 for the testing set).