一种基于两阶段双注意力的溶解氧含量预测模型  

Dissolved Oxygen Prediction Based on Two-Stage Dual-Attention Neural Network

作　　者：周红哲 沈陆明[1] 陈义明 ZHOU Hong-zhe;SHEN Lu-ming;CHENYi-ming(College of Information and Intelligence,Hunan Agricultural University,Changsha 410128,PRC)

机构地区：[1]湖南农业大学信息与智能科学技术学院,湖南长沙410128

出　　处：《湖南农业科学》2022年第2期67-71,共5页Hunan Agricultural Sciences

基　　金：湖南省科技计划重点研发项目(2020NK2033);湖南省教育厅科学研究重点项目(20A249);2020湖南省研究生教育创新工程和专业能力提升工程(CX20200678)。

摘　　要：溶解氧含量是水产养殖水质质量的重要指标,溶解氧含量预测对于水产养殖业的智能化管理与控制具有重大意义。然而,受各种外部相关因素的干扰以及本身不规则变化的影响,溶解氧含量预测精度难以达到理想水平。为了克服上述问题,提出了两阶段双注意力的溶解氧预测模型,该模型可以充分利用溶解氧含量的时间依赖特征和外部相关因素的时空特征,实现更加精准的溶解氧含量预测。通过大量对比试验表明,提出的模型在溶解氧含量预测中获得了当前的最佳性能。Dissolved-oxygen content is an important indicator of water quality in aquaculture,and its prediction is of great significance for the intelligent management and control of the aquaculture industry.However,the accuracy of dissolved-oxygen content prediction is difficult to reach the desired level due to the interference from various external related factors and its irregular variation.To overcome the above problems,a two-stage dual-attention dissolved-oxygen prediction model was proposed,which could make full use of the time-dependent characteristics of dissolved-oxygen content and the spatio-temporal characteristics of external correlation factors to achieve more accurate prediction of dissolved-oxygen.Extensive comparative experiments showed that the proposed model obtained the best current performance in dissolved-oxygen content prediction.

关 键 词：水质 溶解氧预测 时空特征 注意力机制 

分 类 号：X832[环境科学与工程—环境工程]