汪 翔，何吉祥，佘 磊，张 静
（安徽省农业科学院水产研究所，安徽 合肥 230031）

摘要：针对用传统机理建模不能满足水体中亚硝酸盐浓度变化预测的问题，引用非线性自结合的时间序列网络，建立了基于 NAR神经网络的养殖水体亚硝酸盐预测模型。采用2014年6—10月养殖塘口检测的亚硝酸盐的数据建模，建立了用于养殖水体亚硝酸盐模拟的NAR神经网络，并利用2014年11月的观测数据对模型的模拟能力进行了检验。结果显示，建立的养殖水体亚硝酸盐预测模型，可以很好地模拟水体中亚硝酸盐浓度的变化趋势，模拟的绝对误差平均值为0.001 6 mg/L，纳什效率系数为0.72。研究表明，基于NAR神经网络建立的预测模型，在养殖水体亚硝酸盐含量变化预测中具有很强的非线性动态描述能力，对养殖水体中亚硝酸盐的预测有较好的适应性和预测精度。

关键词：养殖水体；水质预测；NAR神经网络；非线性系统

Establishment of aquaculture water nitrite prediction model based on NAR neural network
WANG Xiang，HE Jixiang，SHE Lei，ZHANG Jing
(Fisheries Research Institute, Anhui Academy of Agricultural Sciences,Hefei, 230031 China)

Abstract: In view that the traditional model cannot meet with the prediction of nitrite concentration variations in the water. In the present study, NAR artificial neural network model was developed to predict the variations of water nitrite in aquaculture pond. The nitrite content in the corresponding pond was chosen as output variable. The above data were collected everyday from June to October in 2014 which were used to develop model in this study, and the data collected in November of 2014 were chosen to evaluate the model. The results showed that the changing trend of water nitrite in aquaculture pond could be simulated well by the model, the predictive absolute error mean was 0.0016mg/l, and the efficiency of Nash-Sutcliffe coefficient was 0.72. The prediction model based on NAR neural network had a strong ability to describe the nonlinear dynamic variations of nitrite content in aquaculture water, and it showed the good adaptability and accuracy in practical application.

Key words: aquaculture water; water quality prediction; NAR neural network; nonlinear systems