Machine learning-based optimization of foundation pit dewatering to reduce environmental impact

Abstract

Foundation pit dewatering that combines waterproof curtain and dewatering well is typically adopted to ensure safety in foundation pit engineering, resulting in groundwater level drawdown (ΔH) and ground settlement (Δs) outside the foundation pit. The parameters of the waterproof curtain and dewatering must be optimized to reduce the value of ΔH and Δs. Conventionally, numerical simulations are used for design optimization. However, a substantially long computation time is required to simulate a large number of working conditions. This paper proposes an evolutionary multi-layer neural network (EMNN)-based model that combines the differential evolution algorithm (DEA) and multi-layer neural network (MNN) to optimize the design of foundation pit dewatering, considering ΔH as the control target. Based on the foundation pit dewatering engineering database in Shanghai, an optimization scheme is developed using the EMNN for a dewatering case in Shanghai. Compared with the original scheme, the filter length (L) of the optimal scheme is reduced by 2 m and the vertical position relative to the waterproof curtain (R_p) decreased by 4 m, and ΔH at an observation well outside the pit (5 m away from the waterproof curtain) reduced by 0.08 m. Numerical simulations are employed to calculate Δs due to dewatering. Compared with the original scheme, the maximum value of Δs at the observation well outside the pit and the influence range of ground settlement outside the pit of the optimal scheme are simultaneously reduced.