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

Abstract: 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, 15parameters 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 0.941 3 and 0.996 5,respectively.The values of mean square error were 0.006 5 and 0.063 3,respectively.These results demonstrated that,both models had good predictive performance,which indicated that BP-ANN algorithm was an effective means to establish models of economic benefits and ecological effects of fish ponds.

Key words: BP neural network, Principal component analysis, Fish pond, Fish production, Amount of exchanged water, Economic benefit, Ecological effect