Abstract: There are many channels for oil and gas flow in the underwater oil production tree, and the erosion of the fluid on the wall of the production channel will lead to its failure. The erosion characteristics of the production channel model are studied by means of CFD(computational fluid dynamics) software, and the data obtained from CFD simulation are trained and predicted by BP neural network. Then the BP neural network model is incorporated into genetic algorithm GA to establish the optimization model of the flow channel structure based on GA-BP to maximize the erosion rate. Minimum value is the objective of optimization to optimize the structure of the production channel in order to prolong its service life.