基于ADASYN和Stacking集成的南太平洋黄鳍金枪鱼渔场预报模型研究  

Research on fishing ground forecast models of South Pacific Thunnus albacores based on ADASYN and Stacking integration

作　　者：张聪 周为峰[1] 樊伟[1] ZHANG Cong;ZHOU Weifeng;FAN Wei(East China Sea Fisheries Research Institute,Chinese Academy of Fishery Sciences,Shanghai 200090,China;Graduate School of Chinese Academy of Agricultural Sciences,Beijing 100081,China)

机构地区：[1]中国水产科学研究院东海水产研究所,上海200090 [2]中国农业科学院研究生院,北京100081

出　　处：《海洋渔业》2023年第5期544-558,共15页Marine Fisheries

基　　金：国家重点研发计划(2019YFD0901405);中央级公益性科研院所基本科研业务费(2019T09)。

摘　　要：为提供更为准确的南太平洋黄鳍金枪鱼(Thunnus albacores)渔场预报信息,针对传统渔场分类问题中渔场和非渔场样本数据分布不均衡的问题,提出了一种基于自适应综合过采样(ADASYN)和Stacking集成的渔场分类模型——A-Stacking模型。利用2008—2019年南太平洋黄鳍金枪鱼的渔业数据,结合时空因子、海洋环境因子共32个特征要素(月份、经纬度、海平面异常、涡动能、叶绿素a浓度、叶绿素梯度、叶绿素距平、海表温度梯度、海表温度距平以及0~500 m水层的垂直温度和盐度)建立了南太平洋黄鳍金枪鱼渔场预报模型。为了验证模型的可靠性,将CART、Adaboost、GBDT、XGBoost、KNN、RF和Stacking模型对照。结果显示,A-Stacking模型具有更高的准确率、召回率、F1-score、G-mean和AUC值,且模型的ROC曲线和PR曲线能较好地包含其他模型,表明模型的分类效果更好。研究表明,A-Stacking集成模型对南太平洋黄鳍金枪鱼渔场的预报效果较好,能有效处理不均衡数据的渔场分类问题,可为今后的渔场预报方法提供参考。In order to provide more accurate South Pacific yellowfin tuna(Thunnus albacores)fishery forecast information,and considering the unbalanced distribution of fishery and non-fishery sample data in the traditional fishery classification,this paper proposed a method based on adaptive comprehensive oversampling(ADASYN)and Stacking integrated fishery classification model—A-Stacking model.Firstly,the adaptive comprehensive sampling method was used to oversample the original training data set and principal component analysis was used to reduce the dimension.Secondly,AdaBoost,KNN,CART and RF were used as the base learners of Stacking integration framework,and logical regression(LR)was used as the meta learner to build A-Stacking model.This paper used the fishery data of yellowfin tuna in the South Pacific from 2008 to 2019,combined with space-time factors and marine environmental factors,a total of 32 characteristic elements(month,latitude and longitude,sea level anomaly,eddy kinetic energy,chlorophyll a concentration,chlorophyll gradient,chlorophyll anomaly,sea surface temperature gradient,sea surface temperature anomaly,and vertical temperature and salinity of 0-500 m water layer)were used as the input of A-Stacking model,and the accuracy of A-Stacking model was tested with real fishery production data.In order to verify the reliability of the model,this paper used CART,Adaboost,GBDT,XGBoost,KNN,RF and Stacking models as controls.The comparison models all used the same dataset.All models had optimal parameters determined by grid search.The temporal resolution of the models was month,and the spatial resolution was 0.5°×0.5°.The experimental results showed that compared with the single model,the Stacking integrated model had higher accuracy,recall,F1 score,G-mean value and AUC value,which were 76.80%,74.21%,67.62%,69.79% and 0.8226 respectively,indicating that the classification accuracy of the model was higher than that of the single model.The A-Stacking integration model has further improved on the basis of Stacki

关 键 词：渔场预报 ADASYN STACKING 分类 黄鳍金枪鱼 南太平洋 

分 类 号：S934[农业科学—渔业资源]