基于RNN的水厂过滤过程研究  

Research on the Filtration Process of Water Plants Based on RNN

作　　者：张维玺 洪雷[1] ZHANG Weixi;HONG Lei(School of Environmental and Municipal Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学环境与市政工程学院,甘肃兰州730070

出　　处：《现代信息科技》2024年第14期140-144,共5页Modern Information Technology

摘　　要：当下饮用水标准不断提高,传统水质预测方法存在明显不足。对使用循环神经网络(RNN)处理时间序列数据进行了研究。主要针对全年数据、夏季数据、冬季数据进行分析,结果显示全年数据最优模型MSE和MAE为0.0047、0.0541;冬季数据最优模型MSE和MAE为0.0051、0.0544;夏季数据的模拟效果相对较差,其最低MSE和MAE值为0.2859、0.4704。说明RNN在对大量的水质数据进行预测时,其模拟有效且拟合精度很高,但对数据量少、数据情况复杂的模型模拟时,其拟合效果并不是很好。The current drinking water standards are constantly improving,and traditional water quality prediction methods have obvious shortcomings.A study is conducted on the use of Recurrent Neural Networks(RNNs)for processing time series data.The analysis mainly focuses on annual data,summer data and winter data,and the results show that the optimal model MSE and MAE models for annual data are 0.0047 and 0.0541,respectively.The optimal model MSE and MAE for winter data are 0.0051 and 0.0544,respectively.The simulation effect of summer data is relatively poor,with the lowest MSE and MAE values of 0.2859 and 0.4704.It shows that when RNN is used to predict a large amount of water quality data,its simulation is effective and the fitting accuracy is high.However,when simulating models with small amounts of data and complex data situations,the fitting effect is not very good.

关 键 词：过滤水质预测 循环神经网络 时间序列 最优模型 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]