相模湾西湘地域の定置網における複数魚種の漁獲量予測 An Approach to Neuro-computing for Forecasting Catches of Multiple Species in the Set Net of Seishyo Region, Western Sagami Bay
In this paper the supervised learning paradigm with three layered networks and backpropagation algorithm of artificial neural networks were used for forecasting catches of multiple species in set nets. Three-month means of CPUE of jack mackerel <i>Trachurus japonicus</i>, Japanese sardine <i>Sardinops melanostictus</i>, chub mackerel <i>Scomberjaponicus</i> and total catch for all species in set nets of the Seishyo region in Sagami Bay were collected as output vectors. The input vectors of oceanic conditions were the average temperature anomalies around the set nets and at the surface and 50m depth in Sagami Bay, the Kuroshio path type, and the distances of the Kuroshio current from Cape Irou-zaki and Cape Nojima-zaki. We examined four cases of networks considering fish species and fishing seasons. The CPUE of jack mackerel and total catch in the set nets can be predicted from those of the preceding period and oceanic conditions. Predictions using the same season data were more successful than those using the preceding seasons.
日本水産学会誌 63(4), 549-556, 1997-07-15