南海，海洋环境独特，生物种类繁多，蕴藏着丰富的渔业资源．近年来，近海过度捕捞现象愈发严重，渔业资源已逐渐衰竭，开发外海资源势在必行．鸢乌贼(Sthenoteuthis oualaniensis) 资源量高且营养价值丰富，是南海外海重要的经济头足类鱼种．因此，为合理开发鸢乌贼资源的同时科学指导渔业生产，准确的渔获量预测十分重要．目前鸢乌贼的渔获量数据主要有科研信息船的局部调查数据和渔民自行上报的电子捕捞日志数据．针对以上两种数据，本文展开了如下研究．

（1）针对可靠性较高但数据量较小的信息船调查数据，本文提出了一种新的改进初值和残差的灰色 Verhulst 模型来进行渔获量预测．在新的预测模型中，首先根据新信息优先原理优化模型的初值结构，并结合约束优化模型求解初值参数，使得改进后的初值在重视新信息的同时充分利用各时点信息，显著降低了针对波动性较强的数据序列的预测误差．然后根据傅里叶级数修正模型残差，可有效削弱原始数据中的噪声，改善了灰色模型长期预测时精度衰减的问题．最后实验结果表明，改进初值和残差的优化方法能够有效地提高灰色 Verhulst 模型的模拟和预测精度，在对 4 条信息船的渔获数据的模拟预测中，平均相对误差分别达到了 2.59%、2.06%、1.25%和 0.57%，预测效果优于其他对比模型，验证了优化模型的有效性和可行性．

（2）针对数据量充足但质量相对不佳的电子捕捞日志数据，本文提出了一种基于遗传算法的随机森林 (GA-RF)模型，可进行鸢乌贼渔获量的准确预测．首先，分别通过 BP 神经网络、支持向量机和随机森林等方法建立了渔船捕捞参数与渔获量之间关系的回归模型．通过对上述模型预测结果的对比分析，并根据随机森林模型可直接给出特征重要性的优势，使用遗传算法对随机森林的关键参数进行全局寻优．该方法比依靠经验和网格搜索确定参数的传统方法更加科学有效，且 GA-RF 模型的识别精度更高，在测试集上的拟合优度达 0.905，更具实用价值．最后，基于 GA-RF 模型对渔船捕捞参数进行权重提取，结果如下：船长 (25.76%) 、总吨 (22.06%) 、主机功率 (20.86%) 、灯泡功率 (17.26%) 、平均作业时长 (11.63%) 、平均网次数量(2.43%) ．将各影响因素的权值代入生成的模型中，可进行南海外海鸢乌贼的渔获量预测．

Because of the unique marine environment, the South China Sea is rich in fishery resources. In recent years, due to the increasing offshore overfishing, fishery resources have been gradually depleted, so development in the open South China Sea must be taken seriously. Sthenoteuthis oualaniensis, a high-quality aquatic product rich in many nutrients, is an important economic cephalopod in the open South China Sea. In order to exploit rationally and to guide fishery production scientifically, it's very necessary to assess the catch of Sthenoteuthis oualaniensis. The current fishery data mainly based on few information vessels and electronic fishing logbook reported by fishermen themselves. According to these two different types of data, the main research results are as follows.

(1) The amount of fishery data from information vessels are relatively small, Based on this kind of data, a new grey Verhulst model with improved initial values and residuals was proposed for catch prediction. In the new prediction model, the initial value of the original model was optimized by new information priority principle, and through constrained optimization model, the parameters were solved, so that the improved initial value could make full use of data. Besides, the residual errors were corrected according to the Fourier series, which can effectively weaken the noise in the original data series and the accuracy attenuation problem in the long-term prediction of gray model. The result shows that the optimized method can improve the simulation and prediction accuracy of the grey Verhulst model effectively. In the simulation and prediction of catch data from the four information vessels, the average relative errors reached 2.59%, 2.06%, 1.25% and 0.57% respectively. It shows that the optimized grey Verhulst model is better than other comparison models, which verifies the effectiveness and feasibility of the new model.

(2) Based on the fishery monitoring information with poor quality, this paper proposed an optimized algorithm involving genetic algorithm combined random forest model that can get accurate catch prediction. Firstly, a regression model of fishing parameters and catch data was established by BP neural network, support vector regression and random forest. Secondly, by analyzed the prediction results of these models and accorded to the feature importance scores that random forest model can give, a genetic algorithm was used to perform a global optimization for the key parameters of random forest model, which is more effective than the traditional methods that rely on experience and grid search. The result shows that the GA-RF model has a higher identification accuracy and a better fit of 0.905 on the test set. Finally, the weights were extracted for each indicator, and the results were as follows: boat length (25.76%), gross tonnage (22.06%), main engine power (20.86%), bulb power (17.26%), average operating time (11.63%) and average time of nets (2.43%). By substituting the weights of the influencing factors into the generated model, catch prediction of Sthenoteuthis oualaniensis in the open South China Sea can be made.