Abstract:To accurately assess the damage degree of sudden water pollution in the upstream of the reservoir area, according to the structural characteristics of narrow upstream and wide downstream in the upstream riverbed of the reservoir area, one-dimensional and two-dimensional hydrodynamic and water quality space coupling models were established and improved. Based on the traditional overlapping projection method, a quadratic transverse distribution function of the flow velocity, water level, pollutant concentration and other physical quantities were introduced at the boundary of the overlapping area. Based on the idea of neural network, the steepest descent method was adopted and a training algorithm was constructed to train the unknown parameters in each distribution function. Finally, the converged distribution function was used as the connection condition of the model. In the case of Qinglong River Basin in Taolinkou, the necessary parameters of the distribution function were obtained through training, and it was taken as the connection of one-dimensional and two-dimensional to simulate the temporal and spatial variation law of pollutants in the river environment. The results show that the calculation results of the improved coupling model are stable, which significantly reduces the mutation of physical quantity in the transition from one to two dimensions. Compared with the single numerical transfer of traditional overlapping projection, the improved method takes the trained distribution function as the connection of 1D-2D coupling, which can greatly improve the accuracy of coupling.