Environment-aware and sustainable design of crushed gravel bridge piers based on deep learning monitoring models

In order to solve the nonlinear, parameter-sensitive and environmental instability of bridge piers, a new approach is proposed, which combines a dual-channel residual neural network with feature enhancement. Based on the normalized multidimensional data set, the proposed model captures the ground response in real time, enabling the prediction of settlement, stress, and deformation under various subsurface conditions. Moreover, an improved GA is used to optimize multi-objective sustainable design, which can minimize the use of materials and keep the security constraints. The simulation results show that the settlement error is within ± 1.3 mm, the main stress error is less than 1.8mpa, and the use of concrete decreases 12.5% on average. This approach provides a feasible solution for smart ground monitoring and flexible infrastructure design in complicated geology and environment.