经验预报

Empirical forecasting

依赖个人经验，简单易行

Rely on personal experience， simple and easy

预测精度低，受个人经验限制

Prediction accuracy is low and limited by personal experience

低

Low

统计预报

Statistical forecasting

基于历史数据，简单易行

Based on historical data， simple and easy

依赖历史数据，预测结果有限

Rely on historical data and have limited predictions

中等

Medium

数值模拟与模型预报

Numerical simulation and model forecasting

处理多变量，精确预测

Handle multivariate and forecast accurately

需要大量计算资源

Requires a lot of computing resources

高

High

遥感与地理信息系统

Remote sensing and GIS

实时监测，大范围覆盖

Real-time monitoring， large-scale coverage

受天气影响，数据精度有限

Due to weather conditions， data accuracy is limited

高

High

大数据与人工智能

Big data and artificial intelligence

自动学习，动态预测

Automatic learning， dynamic prediction

需要大量数据和计算资源

Requires a lot of data and computing resources

最高

Highest

综合预报与多学科融合

Integrated forecasting and multidisciplinary integration

多学科知识整合，全面预测

Multidisciplinary knowledge integration for comprehensive forecasting

数据处理复杂，模型构建难度大

Data processing is complex， and model construction is difficult

高

High

广义可加模型

Generalized additive model

能够较好地拟合和预测金枪鱼的分布，能够捕捉非线性关系，模型解释能力强

It can fit and predict the distribution of tuna well， can capture nonlinear relationships， and has strong model interpretation ability

计算复杂度较高，对数据量要求较大

The computational complexity is high and the data volume requirements are large

最高

Highest

广义线性模型

Generalized linear model

结构稳定、操作简单，适用于线性关系明显的数据

The structure is stable， the operation is simple， and it is suitable for data with obvious linear relationship

对数据的线性假设要求较高，实际渔业数据可能不满足，导致效果较差

The linearity assumptions of the data are required， and the actual fishery data may not be satisfied， resulting in poor results

中等偏上

Upper medium

贝叶斯模型

Bayesian model

能够利用先验信息，对不确定性有较好的处理能力

Able to use prior information and have a good ability to deal with uncertainty

需要较多的先验信息，计算复杂度高，模型构建和计算时间较长

It requires a large amount of prior information， high computational complexity， and takes a long time to build and computation

中等

Medium

栖息地指数模型

Habitat suitability index

基于环境因子构建，直观易懂，可解释性强

It is built based on environmental factors， which is intuitive and easy to understand， and has strong interpretability

对环境因子的线性假设较强，难以捕捉复杂的非线性关系，预报精度有限

The linearity assumptions of environmental factors are strong， and it is difficult to capture complex nonlinear relationships， and the prediction accuracy is limited

中等偏下

Medium to low

灰色关联模型

Grey relational analysis

对数据的平滑处理较强，适用于数据量较少的情况

It has strong smoothing of data and is suitable for small amounts of data

难以捕捉细微的变化，对非线性关系的捕捉能力弱，预报精度较低

It is difficult to capture subtle changes， the ability to capture nonlinear relationships is weak， and the prediction accuracy is low

最低

Minimum

黄鳍金枪鱼

Larvae and Juveniles

GAM模型的优点在于其简单易用和可解释性强，适合处理线性或弱非线性关系的数据

The advantages of the GAM model are that it is easy to use and interpretable， and is suitable for processing data with linear or weak nonlinear relationships

GAM模型的缺点主要在于其拟合能力有限，对复杂数据的适应性较差，且对异常值敏感

The disadvantages of the GAM model are mainly due to its limited fitting ability， poor adaptability to complex data， and sensitivity to outliers

GAM模型 （偏差解释率为29.4%，均方误差为 0.149） 表现较为一般，适合对数据拟合要求不高的场景

The GAM model （deviation interpretation rate of 29.4% and mean square error of 0.149） was relatively average， which was suitable for scenarios with low requirements for data fitting

长鳍金枪鱼

Albacore tuna

在处理非线性关系和解释变量影响方面表现出色，适合用于分析环境因子与资源丰度之间的复杂关系

It excels in dealing with nonlinear relationships and the effects of explanatory variables， and is suitable for analyzing complex relationships between environmental factors and resource abundance

主要体现在模型复杂性、数据量和质量要求、变量选择的局限性、预测精度以及对动态变化和经济因素的考虑不足等方面

It is mainly reflected in the complexity of the model， the amount and quality of data， the limitations of variable selection， the prediction accuracy， and the insufficient consideration of dynamic changes and economic factors

总偏差解释率为 46.4%，AIC 值为 132.36，表明模型对 CPUE 的解释能力有限，但能够提供一定的预测能力

The total bias interpretation rate was 46.4% and the AIC value was 132.36， indicating that the model has limited explanatory power for CPUE， but can provide some predictive power

长鳍金枪鱼

Albacore tuna

体现了其非线性建模能力、变量选择与优化、灵活的误差分布假设、平滑函数处理、强解释性、多变量分析能力以及易于实现与扩展等

It reflects its nonlinear modeling ability， variable selection and optimization， flexible error distribution assumptions， smooth function processing， strong interpretability， multivariate analysis ability， and easy implementation and expansion

模型对CPUE总偏差的解释率仅为11.34%，相对较低，这可能与数据年份少、捕捞方式单一和环境要素选择不全面有关

The explanation rate of the model for the total deviation of CPUE is only 11.34%， which is relatively low， which may be related to the small number of data years， single fishing methods and incomplete selection of environmental factors

GAM 模型对长鳍金枪鱼CPUE 总偏差解释率为 11.34% ， 其中贡献最大的为时间 （周）， 贡献率为 6.35%

The GAM model explains 11.34% of the total deviation of CPUE for albacore tuna， with time （week） being the largest contributor， accounting for 6.35%

机器学习

Machine learning

基于数据和算法的自动化学习，模型多样，灵活性高

Automatic learning based on data and algorithms， diverse models and high flexibility

模型多样，适用于中等规模数据集；训练时间较短，易于理解和解释；鲁棒性较强，不易过拟合

The models are diverse and suitable for medium-sized datasets. The training time is short， easy to understand and explain； It is robust and not easy to overfit

对高维复杂数据的处理能力有限；需要较多的调参工作；在处理大规模数据时可能表现不足

Limited ability to process high-dimensional and complex data； it requires a lot of parameter tuning work； it can be underperforming when dealing with large-scale data

深度学习

Deep learning

多层神经网络，自动提取复杂特征，适合大规模数据

Multi-layer neural network， automatic extraction of complex features， suitable for large-scale data

强大的特征提取能力，适合高维度数据；预测精度高，能够处理大规模数据；自动学习数据中的复杂模式

Powerful feature extraction capabilities， suitable for high-dimensional data； high prediction accuracy and ability to process large-scale data； automatically learn complex patterns in your data

需要大量的标注数据进行训练；计算资源需求高，训练时间长；模型解释性较差

A large amount of labeled data is required for training； high demand for computing resources and long training time； the model is poorly interpretable

人工神经网络

Artificial neural network

模拟生物神经网络，非线性拟合能力强

Simulating biological neural networks， with strong nonlinear fitting ability

非线性拟合能力强，能够处理复杂的数据关系；灵活性高，适用于多种数据类型

Strong nonlinear fitting ability， able to deal with complex data relationships； flexibility for a wide range of data types

需要大量数据进行训练；容易过拟合；训练过程复杂，模型解释性差

Requires a lot of data for training； Easy to overfit； the training process is complex and the model is poorly interpretable

XGBRF模型

XGBRF model

XGBRF模型通过结合随机森林和XGBoost的优势，显著提高了模型的预测精度、稳定性和泛化能力，适用于复杂的海洋渔业数据建模和渔场预报任务

By combining the advantages of random forest and XGBoost， the XGBRF model significantly improves the prediction accuracy， stability and generalization ability of the model， which is suitable for complex marine fishery data modeling and fishery forecasting tasks

存在模型复杂度高、数据预处理复杂、解释性不足以及对数据质量和数量要求较高等问题

There are some problems， such as high model complexity， complex data preprocessing， insufficient explanatory ability， and high requirements for data quality and quantity

准确率：75.39%，

渔场召回率：87.36%，

渔场F1得分：82.64%，

非渔场查准率：66.32%，

AUC：79.48%，

accuracy：75.39%，

recall of fishing ground：87.36%，

F1-score of fishing ground：82.64%，

precision of non-fishing ground：66.32%，

AUC：79.48%

随机森林模型

Random forest model

随机森林则以其易用性、泛化能力和稳定性脱颖而出，适合快速建模和特征重要性分析，且对数据的鲁棒性较强

Random forests stand out for their ease of use， generalization ability， and stability， which are suitable for rapid modeling and feature importance analysis， and have strong robustness to data

随机森林（预测准确率：64.8%；Kappa 值：最高为 0.51）在预测长鳍金枪鱼的时空分布时表现最佳，具有最高的准确率和稳定性，能够较好地预测不同等级的渔场分布

Random forest （prediction accuracy： 64.8%； Kappa value： up to 0.51） performed best in predicting the spatial and temporal distribution of albacore tuna， with the highest accuracy and stability， and was able to predict the distribution of different fish grades

CNN-SA-XGBoost 模型

CNN-SA-XGBoost model

特征提取能力强、预测精度高、模型适应性强、模型可扩展性强等

Strong feature extraction ability， high prediction accuracy， strong model adaptability， and strong model scalability

易陷入局部最优解、对初始权重敏感、训练时间长、泛化能力有限等

It is easy to fall into the local optimal solution， sensitive to the initial weight， has a long training time， and has limited generalization ability

CNN-SA-XGBoost 模型预测长鳍金枪鱼资源丰度，均方根误差为 0.486

The CNN-SA-XGBoost model predicted the abundance of albacore tuna with a root mean square error of 0.486

CNN-DNN-CCA-RBF融合模型/GBDT-LR模型融合

CNN-DNN-CCA-RBF fusion model/

GBDT-LR model fusion

2种模型均通过多模型融合的方式，充分发挥了各自子模型的优势，减少了单一模型的局限性

Both models give full play to the advantages of their respective sub-models through multi-model fusion， and reduce the limitations of a single model

CNN-DNN-CCA-RBF模型的主要缺点在于其复杂度高、对数据量要求大以及调参难度大；而GBDT-LR模型则在数据质量依赖性、特征工程依赖性和过拟合风险方面存在挑战

The main disadvantages of the CNN-DNN-CCA-RBF model are its high complexity， large data requirements， and difficulty in parameter tuning. However， the GBDT-LR model has challenges in terms of data quality dependence， feature engineering dependence and overfitting risk

GBDT-LR 模型在所有指标上均优于 CNN-DNN-CCA-RBF 模型，尤其是在中心渔场召回率和总召回率方面，分别高出2.8%和6.2%

The GBDT-LR model outperformed the CNN-DNN-CCA-RBF model in all indicators， especially in the central fishery recall and total recall， which were 2.8% and 6.2% higher， respectively

人工神经网络

Artificial neural network

在生态信息学中展现了强大的灵活性和适应性，能够处理复杂的非线性关系和不确定性数据

It has demonstrated great flexibility and adaptability in ecoinformatics， and is able to handle complex nonlinear relationships and uncertain data

模型复杂、数据需求大、计算成本高和结果解释性不足等

Model complexity， large data requirements， high computational costs and insufficient explanatory results

在可用的时间序列数据中，正确预测的比例达到了61.9%。这表明在该特定案例中，ANNs结合遗传算法的预测模型具有一定的准确性和实用性

Among the available time series data， 61.9 percent of the predictions were correct. This shows that in this particular case， the prediction model of ANNs combined with genetic algorithm has certain accuracy and practicability

多种BP神经网络预报模型

Backpropagation neural network models

BP神经网络在北太平洋柔鱼渔场预报中的应用，充分发挥了其自适应、非线性拟合、容错性和多因子综合处理等优点

The application of BP neural network in the prediction of soft fish fishery in the North Pacific Ocean gives full play to its advantages of self-adaptation， nonlinear fitting， fault tolerance and multi-factor comprehensive processing

模型过于复杂，且研究仅研究了单隐层BP神经网络结构模型，对于其他结构（如多隐层）的神经网络是否能在渔情预报中应用，或者进一步提高预报精度，还需要进一步研究

The model is too complex， and the study only studied the structure model of the single-hidden layer BP neural network， and further research is needed on whether the neural network with other structures （such as multiple hidden layers） can be applied in fishing forecasting， or to further improve the prediction accuracy

经过对13种不同的BP神经网络模型的比较和验证，最优的模型结构为9-7-1，其拟合残差仅为0.008570

After comparing and verifying 13 different BP neural network models， the optimal model structure is 9-7-1， and its fitting residuals are only 0.008570

EMD-BiLSTM模型

EMD-BiLSTM model

该方法有效处理非线性和非平稳数据，充分发挥神经网络优势，降低预测误差，为金枪鱼渔场预报提供新思路，提升渔业资源管理与预测精度

This method effectively processes nonlinear and non-stationary data， gives full play to the advantages of neural networks， reduces prediction errors， provides new ideas for tuna fishery forecasting， and improves the management and prediction accuracy of fishery resources

在使用EMD分解算法对原始数据序列进行分解时，分解结果中存在模态混叠现象，导致分解分量的平稳性并没有得到彻底的改善

When the EMD decomposition algorithm is used to decompose the original data sequence， there is modal aliasing in the decomposition results， which leads to the stationarity of the decomposition components is not completely improved

与Bi-LSTM模型相比，EMD-BiLSTM模型的均方根误差降低了0.053，绝对误差降低了0.018；与BP模型相比，均方根误差降低了0.208，绝对误差降低了0.048

Compared with the Bi-LSTM model， the root mean square error of the EMD-BiLSTM model is reduced by 0.053 and the absolute error is reduced by 0.018， and the root mean square error is reduced by 0.208 and the absolute error is reduced by 0.048 compared with the BP model