西北太平洋秋刀鱼资源丰度预报模型构建比较

摘要

根据1989—2012年西北太平洋秋刀鱼（Cololabis saira）的单位捕捞努力量渔获量(CPUE)以及对应的海洋环境因子数据，包括1-12月各月的Trans-Nino指数（TNI）、1月黑潮区域海表面温度（SST 黑潮）、6月亲潮海表面温度（SST 亲潮），采用BP神经网络预报模型，对西北太平洋秋刀鱼资源丰度进行预测分析。通过10种神经网络模型比较，以及实际CPUE的验证，以拟合残差最小的预报模型作为最优预报模型。研究表明：各月TNI指数、1月黑潮区域海表面温度、6月亲潮海表面温度对西北太平洋秋刀鱼资源丰度影响显著，结构为14-10-1的 BP神经网络模型相对误差仅为0.000681，可作为西北太平洋秋刀鱼资源丰度的预报模型。%Pacific saury(Cololabis saira)is an important oceanic fish, and its abundance index is significantly influenced by marine environment. According to the catch data (Catch per fishing unit, CPUE) of Pacific saury during 1989 to 2012 in the northwestern Pacific Ocean, and the corresponding marine environmental data, including monthly of Trans-Nino index (TNI) from January to December, the sea surface temperature (SST kuroshio) in the Kuroshio area during January and sea surface temperature in the Oyashio area (SSTOyashio) during June, the forecasting model on abundance index for Pacific saury has been built and compared by using BP neural network . Based on comparing of 14 kinds of neural network models, and the actual validation of CPUE, the forecasting model of fitting residual error to a minimum is considered as the optimal prediction model. The studies indicated that monthly TNI index, SST kuroshio, and SSTOyashio has significant effects on abundance index, The BP neural network model with 14-10-1 structure has the lowest relative error of 0.000 681, which can be used as the forecasting model of abundance index for Pacific saury in the northwestern Pacific ocean.