基于ConvLSTM-CNN预测太平洋长鳍金枪鱼时空分布趋势

Prediction of spatial-temporal distribution trend of Pacific albacore tuna based on ConvLSTM-CNN

海洋渔场的变动由空间与环境因子共同驱动,渔场时空演变信息的精准预测是海洋捕捞的关键。本研究利用1995-2018年太平洋海域长鳍金枪鱼(Thunnus alalunga)的渔业生产统计数据,结合同期海洋环境数据包括海表面温度(Sea Surface Temperature,SST)、海表面盐度(Sea Surface Salinity,SSS)、初级生产力(Primary Productivity,PP)和溶解氧浓度(Dissolved Oxygen Concentration,DO),提出了一种融合卷积长短期记忆网络(Convolutional Long Short-Term Memory Networks,ConvLSTM)和卷积神经网络(Convolutional Neural Networks,CNN)的渔场时空分布预测模型。该模型引入特征提取模块,对时空因子进行编码,提取时空特征信息,同时采用CNN提取海洋环境变量的抽象特征,采用ConvLSTM提取渔业数据的高层时空关联信息,最后融合多种特征对渔场时空演变趋势进行预测。结果表明,模型的均方根误差为0.1036,较随机森林、BP神经网络和长短期记忆网络(Long Short Term Memory,LSTM)等传统渔场预报模型的预测误差降低15%~40%,预测的高产渔区与实际作业的高渔获量区匹配度为89%。该研究构建的渔场时空预测模型能够准确地预测出太平洋长鳍金枪鱼的时空分布,为太平洋长鳍金枪鱼的延绳钓渔业提供科学参考依据。

The changes of fishery resources and location are driven by both spatial and environmental factors.Accurate prediction of spatio-temporal evolution information is the key support of pelagic fishery.The study considered fishery production statistics from the Pacific Ocean from 1995 to 2018,and factors the marine environmental data including Sea Surface Temperature(SST),Sea Surface Salinity(SSS),Primary Productivity,PP)and Dissolved Oxygen concentration(DO),and a spatial-temporal distribution prediction model based on convolutional long short-term memory networks(ConvLSTM)and convolutional neural networks(CNN)were proposed.The model introduced the feature extraction module.the spatial and temporal factors are encoded to extract the spatial and temporal characteristics of the high-rise.Secondly,abstract features of marine environmental variables were extracted by CNN,and spatiotemporal features of fishery data were extracted based on ConvLSTM.Finally,high-level spatiotemporal correlation information was fused to predict the spatiotemporal evolution trend of fisheries.The results showed that the root mean square error of the model was 0.1036,which was 15%~40%lower than the prediction error of traditional fishing ground prediction models such as random forest,BP neural network and Long Short Term Memory network(LSTM).The matching degree between the predicted high-yield fishing area and the actual highyield fishing area was 89%.The spatio-temporal prediction model constructed in this study can accurately predict the spatiotemporal distribution of Pacific albacore tuna and provide scientific reference for the longline fishery of Pacific albacore tuna.