基于Transformer的溶解氧概率预测模型  

Dissolved Oxygen Probabilistic Prediction Model Based on Transformer

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

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：《电脑与信息技术》2022年第5期9-12,15,共5页Computer and Information Technology

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

摘　　要：为了实现更加精确的水产养殖溶解氧预测,文章提出一种基于Transformer的溶解氧概率预测模型。概率预测是根据过去时刻的时间序列信息预测未来可能出现结果的概率分布,该模型首先在编码阶段学习过去时刻序列中复杂的模式,如周期性、季节性、趋势和其他影响变量,提取复杂模式的特征信息。然后在解码阶段预测未来所有可能出现结果的概率分布。基于Transformer的溶解氧概率预测模型可以充分学习序列中复杂的时间模式特征,实验结果表明,在真实数据集上的预测评估中,基于Transformer的溶解氧预测模型具有良好的性能表现。To achieve more accurate dissolved oxygen content prediction,a probabilistic model of dissolved oxygen prediction based on Transformer is proposed in this paper.Probabilistic forecasting is to predict the probability distribution of possible future outcomes based on time series information from past moments.The model first learns complex patterns in past moment series,such as periodicity,seasonality,trends,and other influencing variables,at the encoder stage to extract information about the characteristics of complex patterns.The probability distribution of all possible future outcomes is then predicted in the decoder stage,and the model can fully learn the complex temporal pattern features in the sequence.The experimental results show that the Transformer-based dissolved oxygen prediction model has good performance in the prediction evaluation on real data sets.

关 键 词：概率预测 时间序列 注意力机制 溶解氧 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]