CAJ | 학술논문

应用BP人工神经网络模型模拟泾河陕西段氮营养物质NH4＋-N、NO3--N、TN的浓度变化，通过模拟水质参数相关性分析和河流机理性水质模型分析，确定模型的输入因子，从而构建不同结构的BP人工神经网络模型。以实测的月水文、水质、降水量资料进行模型的训练和检验，结果表明：对NO3--N和TN的模拟结果精度较高，不同结构的模拟结果相对误差在18％以内；对NH4＋-N的模拟结果相对误差大，NH4＋-N主要来源于点源排放，年内不稳定排放是模型模拟误差大的主要原因；单参数输出结构的网络模拟结果优于三参数输出结构的，单隐含层结构的BP网络模拟结果要比多隐含层结构的模拟结果更精确。
응용BP인공신경망락모형모의경하협서단담영양물질NH4＋-N、NO3--N、TN적농도변화，통과모의수질삼수상관성분석화하류궤이성수질모형분석，학정모형적수입인자，종이구건불동결구적BP인공신경망락모형。이실측적월수문、수질、강수량자료진행모형적훈련화검험，결과표명：대NO3--N화TN적모의결과정도교고，불동결구적모의결과상대오차재18％이내；대NH4＋-N적모의결과상대오차대，NH4＋-N주요래원우점원배방，년내불은정배방시모형모의오차대적주요원인；단삼수수출결구적망락모의결과우우삼삼수수출결구적，단은함층결구적BP망락모의결과요비다은함층결구적모의결과경정학。
River water quality simulation provides theory evidence for pollutants variation in water environment. By using BP algorithm of ANN,the paper simulated the concentration of ammonia nitrogen,nitrate nitrogen and total nitrogen in Jinghe River,Shaanxi Province. The input parameters of BP network model were selected through simultaneous analysis on correlation between the output parameters and water quality parameters and mechanism modeling of river water quality. The BP network models with different structures were built based on the input and output parameters which were mainly decided by the studied problems. The models were trained and validated using monthly observed data of hydrology,water quality and precipitation. The results show that the BP models of nitrate nitrogen and total nitrogen can precisely simulate the variation of water quality and the errors of water quality simulation with different structures are less than 18%;the models of ammonia nitrogen produce big errors which are prob-ably related to the source of ammonia nitrogen. Ammonia nitrogen comes from the point source discharge and has instability discharge all over three years to cause the big simulation errors. At the same time the simulation results show that the precisions of the models simulating one water quality parameter exceed the ones simulating three parameters;the BP models with one hidden layer exceed the ones with two and three hidden layers in precision.