PSO-MLP模型预测降雨对府河氨氮的影响

PSO-MLP Model Predicts the Effect of Rainfall on Ammonia Nitrogen in Fuhe River

为研究白洋淀上游的保定市区降雨径流对府河水质的影响,采用府河2019年、2020年的常规水质监测数据,基于粒子群算法(particle swarm optimization, PSO)和多层感知机(multi-Layer perceptron, MLP)建立PSO-MLP水质预测模型。分别使用PSO-MLP、MLP、一维水质模型进行对比预测。结果表明:与一维水质模型相比,PSO-MLP平均绝对误差减少64.5%~74.7%;与MLP模型相比,PSO-MLP平均绝对误差减少6.6%~12.6%。选取2021年7月的一次典型降雨对府河膳马庙、安州和南刘庄3个控制断面进行预测,表明PSO-MLP泛化能力更强,预测误差更小,优于一维水质模型和MLP模型。所建立的府河PSO-MLP水质预测模型,可以提前4 h准确预测府河各断面氨氮浓度,平均绝对误差小于0.3 mg/L,可应用于保定市区降雨径流对府河水质污染的预测预警,避免降雨径流通过府河影响白洋淀水质。

In order to study the influence of rainfall runoff in Baoding City on water quality of Fuhe River,based on the conventional water quality monitoring data of Fuhe River in 2019 and 2020,based on particle swarm optimization algorithm(PSO)and multi-layer perceptrons(MLP),the PSO-MLP water quality prediction model was established.PSO-MLP,MLP and one-dimensional water quality model were used for comparison and prediction respectively.The results show that the mean absolute error of PSO-MLP is decreased by 64.5%~74.7%compared with the one-dimensional water quality model.Compared with the MLP model,the mean absolute error is decreased by 6.6%~12.6%.A typical rainfall in July 2021 was selected to forecast the three control sections of Shanmamiao,Anzhou and Nanliuzhuang,which shows that PSO-MLP has stronger generalization ability and smaller prediction error,and is superior to the one-dimensional water quality model and the MLP model.The established PSO-MLP water quality prediction model of Fuhe River can accurately predict the ammonia nitrogen concentration of each section of Fuhe River 4 h in advance,with an average absolute error less than 0.3 mg/L.It can be applied to the prediction and early warning of water pollution of Fuhe River by rainfall and runoff in Baoding urban area,and avoid rainfall runoff through the Fuhe River to affect the water quality of Baiyangdian.