Abstract: ［Objective］ To enhance the safety of metro station foundation pit construction, it is essential to accurately determine the groundwater hydrogeological parameters, improving the accuracy of permeability and storage coefficient in confined aquifers particularly. This is a critical prerequisite for formulating dewatering schemes. Therefore, a more in-depth study of the hydrological parameters of confined aquifers in metro foundation pits is required. ［Method］ Based on the dewatering test at Sports Center Station on Ningbo Rail Transit Line 7, a three-dimensional transient groundwater seepage model is developed with Modflow6 software called via the Flopy module in Python language. An LSTM (long- and short-term memory) deep learning model is introduced to build a surrogate model of confined aquifer water level variations. Combined with a particle swarm optimization algorithm and based on field-measured data, an inverse analysis of the confined aquifer permeability and storage coefficients is conducted. Thereby a method for hydrogeological parameter inversion in metro foundation pits based on surrogate modeling and optimization algorithms is proposed. ［Result & Conclusion］ The obtained inverted vertical permeability coefficient is 0.76×10-5 m/s, the horizontal permeability coefficient is 1.38×10-5 m/s, and the storage coefficient is 6.42×10-5 m-1. When these parameters are input into the numerical seepage model, the calculated data closely matches the measured data in all process, including the stage of rapid water level drop in the initial pumping, the stage of gradual change during the stabilization period, and the stage of water level gradual recovery after pumping stops, validating the feasibility of the inversion method. The use of deep learning-based surrogate modeling combined with optimization algorithms enables efficient and accurate inversion analysis of groundwater parameters.