基于灰色系统的秘鲁鳀资源量预测模型的构建

秘鲁鳀(Engraulis ringens)是栖息于东南太平洋的小型中上层鱼类,是鱼粉的主要来源。有效预测秘鲁鳀的资源量,综合评价环境因子对秘鲁鳀资源量的影响将有利于我国鱼粉进口企业利用环境和气候变化把握秘鲁鳀的生产行情。利用2004—2013年秘鲁鳀的资源量数据,首先对影响秘鲁鳀资源量的环境和气候因子进行灰色关联分析,并基于该结果利用灰色预测模型GM(0,N)模型]进行秘鲁鳀资源量预测模型的构建,同时通过去除某个环境因子的模型与包含所有环境因子的模型的预报精度比较,对环境因子的重要性进行评价。结果表明,包含所有因子(渔场水温(Fishing ground temperature,FGT)、(渔场水温距平(Fishing ground temperature anomaly,FGTA)、南方涛动指数(Southern Oscillation Index,SOI)、Nino 1+2区海表面温度和太平洋年代际振荡指数(Pacific Decadal Oscillation Index,PDOI))的模型1对拟合资源量与实际资源量的平均相对误差为0.197,两个序列间的相关系数为0.544,对验证数据的相对误差为0.434;对比去掉其中某一个因子的模型2-模型6,去掉PDO的模型4效果最好,拟合资源量与实际资源量的平均相对误差为0.187,两个序列间的相关系数为0.663,对验证数据的相对误差为0.274,该模型能够更好地提高模型精度,可作为预测秘鲁鳀资源量的最优模型。

The construction of biomass forecasting model for the anchoveta (Engraulis ringens)by the grey system model

Anchoveta (Engraulis ringens) is a kind of small pelagic fish living in the Southeast Pacific Ocean. It is also an important source of fishmeal. Predicting the anchoveta biomass effectively and evaluating their relationship with environment factors could benefit companies which import the Peruvian fishmeal. Therefore, this study firstly used the grey correlation analysis to analyze the connection between the anchoveta biomass and environmental factors from 2004 to 2013. And then based on these results, we used the grey forecasting modelGM (0, N) model] to build the anchoveta biomass forecasting model. In addition, by comparing between removing a certain environmental factor and containing all the factors to the model, we evaluated the importance of environmental factors. Results showed that the model which contained all the factors(factors including Fishing ground temperature, FGT, Fishing ground temperature anomaly, FGTA, Southern Oscillation Index, SOI, the sea surface temperatureat Nino 1+2 region, Nino 1+2 and Pacific Decadal Oscillation Index, PDO) had the mean relative error of 0.197 between fitting biomass sequence and predicted biomass sequence; the coefficient correlation index between these two sequences was 0.544; the relative error of the validation data is 0.434. Comparing the models from model 2 to model 6 which removed one environmental factor, the model 4 which removed PDOI had the best result:the mean relative error of fitting biomass sequence and predicted biomass sequence was 0.187, the coefficient correlation index between these two sequences was 0.663, and the relative error of the validation data was 0.274. The results indicated that model 4 can improve the accuracy of forecasting model and could be set as the optimal model for predicting the anchoveta biomass.