基于灰色关联度与BP神经网络的清河水库总氮浓度预测模型

为了实现清河水库总氮浓度的预测,建立了基于灰色关联分析的BP神经网络水质预测模型(GRABP),即采用灰色关联度的方法,选取总磷、挥发酚、化学需氧量、pH值、氨氮五项水质指标作为BP神经网络总氮预测模型的输入变量,根据五项最优影响因子与总氮浓度的对应关系,对模型进行了训练,并将训练好的模型应用于2016年8~12月的总氮浓度预测中。结果表明,GRA-BP网络模型较BP网络具有较高的预测精度,预测的相对误差均在5%以内,可为清河水库的水质管理提供科学的指导。

Prediction Model of Total Nitrogen Concentration in Qinghe Reservoir Based on Grey Relational Grade and BP Neural Network

The method based grey correlation analysis was adopted to predict the total nitrogen concentration in Qinghe Reservoir. Total phosphorus, volatile phenol, chemical oxygen demand, pH value and ammonia nitrogen were selected as the input variables of BP neural network based total nitrogen prediction model. The model was trained based on the correspondence relationship between the five optimal factors and the concentration of total nitrogen. The model was applied to the total nitrogen concentration prediction in 2016 from August to December. The results show that the GRA-BP network model has higher prediction accuracy than the BP network model, and the relative error of prediction is less than 5%, which can provide scientific guidance for Qinghe Reservoir water quality management.