基于LSTM-CMSY模型的太平洋蓝鳍金枪鱼资源量数据评估OA

To address the three core challenges confronting Pacific Thunnus orientalis stock assessment-non-uniform international data statistical standards,strong dependence of traditional models on prior distributions of key parameters,and temporal discontinuity issues in China's distant-water fishery data-this study innovatively proposes an LSTM-CMSY assessment framework.By integrating bidirectional long short-term memory(Bi-LSTM)networks with the conventional Catch-MSY model,the framework employs deep learning to estimate complex mapping relationships between environmental carrying capacity coefficient K and fishing pressure within catch time series,substantially reducing the demand for historical biological parameters.This framework innovatively combines temporal feature extraction with population dynamics mechanisms,using attention mechanisms to parse lagged correlations,there by achieving adaptive imputation for missing data and intelligent parameter estimation.Validation results demonstrate that across 96%of simulation scenarios,population parameters predicted by LSTM-CMSY(e.g.,the maximum sustainable yield,MSY)exhibit high consistency with traditional catch-MSY outcomes,with prediction errors falling below 5%in 61.9%of years and only one year(4.8%)showing MAPE between 10%-20%.Notably,even under extreme conditions where data completeness is below 70%,the model maintains prediction accuracy within acceptable ranges(MAPE<15%),demonstrating excellent robustness.The model maintains stability under data gaps less than 6 months,proving its effectiveness in overcoming limitations posed by missing data.This research applies AI-based time series analysis to transoceanic tuna species assessment,not only providing technical support for fulfilling international fisheries management responsibilities,but also establishing a methodological foundation for China's participation in global marine resource governance and enhancing its discourse power in distant-water fisheries,providing a reliable technical pathway for science-based management under data-limited conditions.