Terrigenous pollutants discharge is one of the most important influencing factor which affects nearshore water ecological environment quality. An optimal sewage reduction method for marine pollution sources was proposed by coupling the marine water quality model and neural network algorithm of data driven model. Based on the mathematical calculation of the marine water quality model for the water quality of pollution sources designed working conditions, the concentrations of pollutants in the marine internal gauge stations were obtained. The nonlinear relationship between the state variables (the concentration of pollutants in the marine internal gauge stations) and the control variables (pollution sources) was constructed by the data driven artificial neural network algorithm . The permitted discharge amounts of every pollution source were calculated with the input of the environmental target data in the marine internal gauge stations. Finally, the reduction amount was computed by the amount of actual sewage combining with the permitted discharge. A case study of inorganic nitrogen emission reduction studies for the 4 sea pollution sources in the Xuwei marine district of Lianyungang was used for validating the proposed method. The results showed that, the data driven neural network method has advantages of being nonlinear, concise and flexible and can support basic data for nearshore water pollution control work. At the same time, in the study to use the method of partition emission reductions in different regions can reflect the basic principles of fairness, and optimize the sea pollution emission reduction work.