基于神经网络的规则提取及其渔业应用研究

Extracting rules based on neural network and its application in fisheries forecasting

为了解决远洋渔业中过度依赖经验而产生的盲目捕捞问题,结合海洋环境数据和历史产量数据对渔场进行有效分析,提出了一种基于径向基函数神经网络(Radial basis function neural network,RBFNN)的栖息地指数(HSI)预测方法,并将其应用于印度洋海域大眼金枪鱼(Thunnus obesus)栖息地指数的预测。在RBFNN训练过程中使用模糊C均值(Fuzzy c-means,FCM)聚类算法,在基于神经网络的规则提取过程中首次采用了和声搜索(Harmony search,HS)算法。实验研究表明,利用FCM改进后的RBFNN,均方误差(Mean square error,MSE)达到0.021 6。和声搜索由于算法简单,易于实现,能够应用于训练后的FCM-RBFNN提取分类规则,提取出的规则能够反映该渔业现状。

In order to solve the issue of blind fishing, which arises from over-reliance on experience in offshore fishing, marine environmental and historical production data have been used to effectively analyze the fishery. This method was proposed to forecast indices of the Indian Ocean big eye tuna＇s （Thunnus obesus） habitat based on radial basis function neural network （RBFNN）. Fuzzy c-means clustering algorithm was utilized during training the neural network. While in the process of rule extraction, a harmony search algorithm was used to extract fishery rules from the trained RBFNN. Finally, the proposed method was used to forecast the fishery habitat indices of the Indian Ocean big eye tuna. Experiments showed that harmony search algorithm can extract classification rules from the trained neural network. The extracted rules reflected the status of the Indian Ocean big eye tuna fishery.