基于LSTM-Attention的水质参数预测研究  

Research on Water Quality Parameter Prediction Based on LSTM-Attention

作　　者：梁冰 田斌[1] 洪汉玉[1] LIANG Bing;TIAN Bin;HONG Han-yu(School of Electrical and Information,Wuhan Institute of Technology,Wuhan 430205,China)

机构地区：[1]武汉工程大学电气信息学院,武汉430205

出　　处：《自动化与仪表》2022年第3期80-84,共5页Automation & Instrumentation

摘　　要：水质参数预测是当今绿色发展、生态修复的重要一环,但传统水质预测模型准确度低、泛化能力弱、计算复杂耗时,难以满足大数据时代下水质预测的需求。该文以江西赣州禾丰盆地的水质参数作为研究对象,根据水质参数周期性、非线性以及长时依赖的特征,提出一种结合注意力机制(Attention)和长短期记忆网络(LSTM)的模型。实验结果表明,该文提出的模型在多种评价指标下均优于循环神经网络(SimpleRNN)和LSTM,能够有效预测未来水质参数的变化趋势,具有较强的泛化性能。Water quality parameter prediction is an important part of today’s green development and ecological restoration.However,traditional water quality prediction models have low accuracy,weak generalization ability,and complex and time-consuming calculations,which are difficult to meet the needs of water quality prediction in the era of big data.This paper takes the water quality parameters of the Hefeng Basin in Ganzhou city,Jiangxi province as the research object.According to the characteristics of the periodicity,nonlinearity and long-term dependence of the water quality parameters,this paper proposes a combination of attention mechanism(Attention)and long short-term memory network(LSTM)model.Experimental results show that the model proposed in this paper is better than SimpleRNN and LSTM under a variety of evaluation indicators,can effectively predict the changing trend of future water quality parameters,and has strong generalization performance.

关 键 词：水质参数预测 注意力机制 长短期记忆网络 

分 类 号：TN98[电子电信—信息与通信工程]