The traditional method of determining the relationship between water level and discharge is mainly based on physical law and empirical formula modeling, the accuracy of the model is limited, and the accuracy of the result is often uncertain due to personal experience and level. Machine learning methods can automatically learn the relationship between water level and flow through training models, which has flexibility and adaptability, but lacks interpretation and understanding of physical laws, and requires high data quality and quantity. This paper proposes a method to determine water level and discharge by combining the physical law of water flow with artificial intelligence algorithm. By combining the water level of hydrological stations with the drop of upstream and downstream stations and machine learning model, a flow calculation scheme is constructed to determine the stage-discharge relation. The experimental results of MLP model and LSTM model are compared in Hankou Station. The results show that the stage-discharge relation determination method combined with the physical law of water flow and artificial intelligence algorithm can accurately calculate the flow rate and draw the water level and discharge relationship curve. Meanwhile, the accuracy of LSTM model calculation results is high, and the R2 of the model calculation results of Hankou Station from 2018 to 2020 is above 0.9950, and the MAPE is less than 3%.