结合灰色理论的人工神经网络方法在水质预测中的应用

Study on the prediction of water quality based on artificial neural network combined with grey theory

针对现阶段水质监测中存在的水质变化响应滞后问题,提出了采用灰色预测法、人工神经网络(BP神经网络、径向基神经网络、广义回归神经网络)以及两者组合的方法对水质动态预测进行研究。以太湖流域嘉兴斜路港监测断面为例,并依据后验差检验比值c及小概率精度p对模型预测效果进行了分析。结果表明,对年内预测,通过广义回归神经网络的动态预测值平均相对误差为0.61%,后验差检验比值小于0.65,小误差概率大于0.7;采用灰色结合广义回归神经网络的方法对水质pH值进行预测,平均相对误差仅有0.85%,后验差检验比值小于0.65,小误差概率等于1。研究结果还表明,对年际预测,灰色结合BP神经网络和灰色结合径向基函数神经网络的动态预测值平均相对误差分别为0.57%和0.80%,其后验差比值都小于0.5,小概率误差都为0.9,大于0.8。

In this paper,the grey theory,artificial neural network(back-propagation neural network,radial basis function neural network,and generalized regression neural network),and the combination of these two methods was proposed to study the dynamic prediction of water quality.Taking the Xielugang in Jiaxin as an example,the model prediction effect was analyzed based on the posterior difference test ratio(c)and small probability accuracy(p).The results showed that within a prediction year,the average relative error of the dynamic prediction value of the generalized regression neural network was 0.61%,and the c was less than 0.65,while the p was greater than 0.7,respectively.The results exhibited that the prediction value using the combination of the grey theory and generalized regression neural network,the averaged relative error was 0.85%,the c<0.65,and the p=1.0,respectively.The inter-year prediction based on the combination of the grey theory with BP neural network and radial basis function neural network,the averaged relative error was 0.57%and 0.80,respectively,and the ratio of posterior error was less than 0.5,and the small probability error was 0.9,but greater than 0.8.