基于5参数GA-BP模型的出水水质预测——以宁夏某水厂为例

Prediction of effluent water quality based on 5-parameter GA-BP model——A case study of certain water plant in Ningxia

为解决饮用水处理过程中关键水质参数浊度取样检测时滞性大和精度低的问题,提出了基于遗传算法优化BP神经网络(GA-BP)的出水浊度预测模型。利用2019—2021年银川市某水厂的实测出水浊度及相关水质数据,采用灰色关联度分析对影响出水浊度的输入指标进行筛选,结合Q型聚类分析将样本数据划分为具有不同特征的3类,构建了基于GA-BP神经网络的机器学习模型对出水浊度进行预测,并与传统BP和未分类的预测结果进行对比。结果表明:与未分类相比,利用Q型聚类分析后预测模型的误差评价指标决定系数(R2)和均方根误差(RMSE)分别优化了2.9%和22%;与传统BP神经网络相比,经遗传算法优化后的预测模型误差评价指标R2和RMSE分别优化了2.4%和12%。研究表明,Q型聚类分析和遗传算法均能提高BP神经网络预测模型的泛化能力,减小误差。

To solve the problems of large time delay and low precision in the sampling and detection of key water quality parameters in drinking water treatment,a turbidity prediction model based on genetic algorithm optimized BP neural network(GABP)is proposed.Using the measured turbidity and related water quality data of a waterworks in Yinchuan City from 2019 to 2021,the input indicators that affect the turbidity are screened using gray correlation analysis,and the sample data is clustered into three different groups with different characteristics using Q-type clustering analysis.A machine learning model based on GA-BP neural network is constructed to predict the turbidity,and the results are compared with those of traditional BP and unclassified prediction models.The results show that compared with unclassified prediction,the error evaluation indexes of the prediction model after Qtype clustering analysis are improved by 2.9%and 22%in terms of R2 and RMSE,respectively;compared with the traditional BP neural network,the error evaluation indexes of the prediction model optimized by genetic algorithm are improved by 2.4%and 12%in terms of R2 and RMSE,respectively.The research shows that both Q-type clustering analysis and genetic algorithm can improve the generalization ability of BP neural network prediction model and reduce errors.