With the construction of urban subway network in our country, more and more tunnels will inevitably cross the underwater karst area. Restricted by the complexity of karst stratum and many uncertainties of stratum after grouting reinforcement, the construction risk of shield crossing such stratum is high, while the selection of reasonable shield tunneling parameters is the key to ensure the safety and efficiency of shield tunneling. Based on the project of Changsha Metro Line 3 shield passing through the underwater karst section, firstly, the karst distribution characteristics were defined by statistics and analysis of drilling data. Secondly, by inputting the stratum characteristic parameters and tunnel characteristic parameters, the BP neural network was established which can output the shield tunneling speed, thrust, cutter head torque, excavation chamber pressure, air cushion chamber pressure and synchronous grouting amount. Finally, the sample data was trained and successfully applied to engineering practice. The results show that the output value of the training is in good agreement with the expected value, and the BP neural network model has good adaptability; the output prediction results can effectively reflect the change trend of the actual shield tunneling parameters, and the average error between the predicted value and the actual expected value is less than 13%, which is within the acceptable error range. The results of field application show that the ground settlement is within the safe range, no engineering accident occurs in the process of shield tunneling, the selection of shield tunneling parameters is reasonable, and the attitude control is good. The research results can be used to guide the construction of underwater karst shield tunnel, and the proposed method also provides a new idea for the reasonable selection of shield tunneling parameters in other complex strata.