As a migratory fish in the middle and upper ocean, Thunnus alalunga has become one of the main fishing targets in fishing countries because of its high economic value and wide distribution range. Based on the background of Thunnus alalunga fishery in the South Pacific, this paper proposes a new yield prediction method for fishery applications. According to the longline fishery data of Thunnus alalunga, spatial factors and sea surface temperature, sea surface height and chlorophyll a mass concentration in the South Pacific from 2000 to 2015, the extension neural network model was used to predict the yield of tuna, and Particle Swarm Optimization (PSO) was used for weight optimization. The results show that the total recall rate reaches 68%, which is higher than that predicated with the traditional method, and it has a great advantage in the prediction of high-yield areas where the recall rate reaches 74.2%, but the prediction effect on the middle-yield areas is significantly poorer than that in the high-yield areas and the low-yield areas. The research shows that the particle swarm extension method can solve the problem that the classical domain in the extension neural network is difficult to determine, and it has a certain guiding effect on enriching the fishery prediction method and reasonable fishing operation.