采用主成分分析和BP神经网络算法预测池塘养鱼产量和换水量  

Prediction of Fish Production and Amount of Exchanged Water of Aquaculture Pond by Principal Component Analysis and BP Neural Network Algorithm

作　　者：张俊彪 陈雯 蔡春芳[1] 何捷 ZHANG Junbiao;CHEN Wen;CAI Chunfang;HE Jie(School of Biology & Basic Medical Science,Soochow University,Suzhou 215123,China;Changzhou University,Changzhou 213164,China)

机构地区：[1]苏州大学基础医学与生物科学学院,江苏苏州215123 [2]常州大学,江苏常州213164

出　　处：《河南农业科学》2020年第6期174-180,共7页Journal of Henan Agricultural Sciences

基　　金：“十二五”农村领域国家科技计划课题(2015BAD13B00)。

摘　　要：产量和换水量分别是池塘养鱼经济效益和生态影响的重要衡量指标。为探讨影响常规鱼类养殖产量和换水量的主要因子,建立基于反向传播(BP)神经网络(ANN)算法的预测模型,通过调查获得51组关于混养草鱼(Ctenopharyngodon idellus)、鲫鱼(Carassius auratus)、鲤鱼(Cyprinus carpio)池塘的完整管理信息。经主成分分析(PCA),草鱼放养密度、鲫鱼放养密度、鲫鱼放养规格、鲤鱼捕捞规格、鳙鱼(Aristichthys nobilis)捕捞规格、鲤鱼产量、塘租费、苗种费、饲料费、电费、调水费、病害防治费、人工费、水深及是否发生蓝藻等15个参数均被筛选入放养鱼类总产量和夏季换水量模型中。放养鱼类总产量模型中还筛选进鲫鱼产量、鳙鱼产量和增氧方式3个参数。夏季换水量模型中还筛选进鲤鱼放养密度和鲢(Hypophthalmichthys molitrix)、鳙放养时间2个参数。随机选取45组数据采用BP-ANN算法建模并预测另外6组数据。结果显示,放养鱼类总产量模型相对误差(RE)最大为8.40%,绝对误差(AE)最大为2.53 t/hm2,平均相对误差(MRE)为5.81%,平均绝对误差(MAE)为1.51 t/hm2。夏季换水量模型AE值最大为19.10 cm,MAE值为13.36 cm。2种模型决定系数(R2)分别是0.9413、0.9965,均方误差(MSE)分别是0.0065和0.0633。总体拟合性能良好,表明BP-ANN是建立养鱼池塘经济效益和生态影响模型的有效手段。The fish production and exchanged water amount of aquaculture ponds are important indexes to evaluate their economic benefits and ecological effects,respectively.To explore the main factors affecting the fish production and the amount of exchanged water of conventional carp ponds,and to build forecasting models based on back propagation neural network(BP-ANN)algorithm,51 complete managemental records about the polyculture ponds of grass carp(Ctenopharyngodon idellus),crucian carp(Carassius auratus)and common carp(Cyprinus carpio)were obtained through survey.PCA showed that,15 parameters including stocking density of grass carp and crucian carp,size of stocked crucian carp,size of harvested common carp and bighead carp(Aristichthys nobilis),production of common carp,cost of pond renting,fingerlings,feed,electric power,material for water quality regulation,costs for disease prevention and salary,water depth,occurrence of algal bloom were screened into both models of the total production of stocking fish and the amount of exchanged water in summer.The production of crucian carp and bighead carp,and oxygenation mode were also screened into the model of the total production of stocking fish.The stocking density of common carp and the stocking time of silver carp(Hypophthalmichthys molitrix)and bighead carp were also screened into the model of the amount of exchanged water in summer.45 records were randomly selected to model by BP-ANN algorithm and the other 6 records were used to predict.The results showed that the values of maximum relative error,maximum absolute error,mean relative error and mean absolute error of the model of total production of stocking fish were 8.40%,2.53 t/ha,5.81%,1.51 t/ha,respectively.The values of maximum absolute error and mean absolute error of the model of the amount of exchanged water in summer were 19.10,13.36 cm,respectively.The values of R2 of the two models were 0.9413 and 0.9965,respectively.The values of mean square error were 0.0065 and 0.0633,respectively.These results demonstra

关 键 词：BP神经网络 主成分分析 养鱼池塘 养鱼产量 换水量 经济效益 生态影响 

分 类 号：S964.3[农业科学—水产养殖]