摘要
水质预测对水资源管理及水体保护至关重要,为提高污水水质预测模型准确率,考虑到水质参数是一个动态的时间序列,在研究RNN神经网络模型基础上,引入一种改进的长—短记忆网络结构(LSTM-GRU)来增加RNN的隐层,GRU和LSTM采用门结构代替标准RNN结构中的隐藏单元,可以选择性地记忆重要信息而忘记不重要信息,从而高效学习历史水质参数信息,使得预测结果更加精确。通过仿真分析,本文采用的LSTM-GRU模型与传统的污水水质参数预测模型相比,LSTM-GRU模型的泛化能力更强,预测精度更高,有效性及实用性更强。
Water quality prediction is very important to water resources management and water body protection.In order to improve the accuracy of sewage water quality prediction model,considering that water quality parameters were a dynamic time series,based on the research of RNN neural network model,an improved long-short was introduced.Memory network structure(LSTM-GRU)was used to increase the hidden layer of RNN.GRU and LSTM used gate structure to replace the hidden unit in the standard RNN structure,which could selectively memorize important information but forget unimportant information,thereby efficiently learning historical water quality parameter information,which made the prediction results more accurate.Through simulation analysis,the LSTM-GRU model used in this paper was compared with the traditional sewage water quality parameter prediction model.The LSTM-GRU model had stronger generalization ability,higher prediction accuracy,and stronger validity and practicability.
作者
邹可可
李中原
穆小玲
李铁生
于福荣
Zou Keke;Li Zhongyuan;Mu Xiaoling;Li Tiesheng;Yu Furong(Pingdingshan Hydrology and Water Resources Survey Bureau of Henan Province,Pingdingshan 467000,China;Hydrology and Water Resources Bureau of Henan Province,Zhengzhou 450003,China;Zhengzhou Hydrology and Water Resources Survey Bureau of Henan Province,Zhengzhou 450006,China;College of Chemistry and Molecular Engineering,Zhengzhou University,Zhengzhou 450052,China;North China University of Water Resources and Electric Power,Zhengzhou 450046,China)
出处
《能源与环保》
2021年第12期59-63,共5页
CHINA ENERGY AND ENVIRONMENTAL PROTECTION
基金
国家自然科学基金资助项目(41402225)
2020年度河南省水利科技攻关项目(GG202038)。
关键词
水质预测
神经网络
长—短记忆模型
门控循环单元
water quality prediction
neural network
Long Short-Term Memory(LSTM)
Gate Recurrent Unit(GRU)