基于GAT-Transformer时间序列模型的SOWQP
Research on Water Quality Prediction Based on Gat-Transformer Time Series Model
王永生 1陈振 1刘利民 1刘广文1
作者信息
- 1. 内蒙古工业大学数据科学与应用学院,内蒙古 呼和浩特 010080;内蒙古自治区基于大数据的软件服务工程技术研究中心,内蒙古 呼和浩特 010080
- 折叠
摘要
水质污染现象日趋严重,水质预测对于水质的保护尤为重要.然而水质受到多种水质指标的影响,并且水质监测数据是一种时序数据,各种指标间存在着复杂关系,现有的方法并不能充分捕捉水中各指标间的复杂关系.针对当前水质时序数据预测长距离依赖和多种指标间关系考虑不充分的问题,提出一种基于GAT-Transformer的水质预测方法.首先引入卷积神经网络(CNN)提取每个水质时间序列数据输入的高级特征;然后将CNN输出的数据输入GAT层和多头自注意力层,通过GAT捕捉水中各元素间的复杂关系;最后将GAT层和多头自注意力层的输出结合,通过Transformer进行预测输出.使用内蒙古某真实水质数据集进行实验,表明所提方法在MAE、MSE和RMSE三个指标上的综合表现优于其它方法.
Abstract
The phenomenon of water pollution is becoming increasingly serious,and water quality prediction is particularly important for water quality protection.However,water quality is influenced by multiple water quality indi-cators,and water quality monitoring data is a temporal data with complex relationships between various indicators.Ex-isting methods cannot fully capture the complex relationships between various indicators in water.A water quality pre-diction method based on gat transformer is proposed to solve the problems of long-distance dependence of current wa-ter quality time series data prediction and insufficient consideration of the relationship between various indicators.First,convolutional neural network(CNN)was introduced to extract the advanced features of each water quality time series data input;Then the data output from CNN was input into the gat layer and the multi head attention layer,and capture the complex relationships among the elements in the water through the gat;Finally,the output of GAT layer and multi head attention layer were combined to predict the output through transformer.In this study,a real water quality data set in Inner Mongolia was used for experiments,which shows that the comprehensive performance of this method in Mae,MSE and RMSE,is better than other methods.
关键词
图注意力/水质预测/多头自注意力Key words
Graph attention/Research on water quality prediction(SOWQP)/Multi-Head Attention引用本文复制引用
基金项目
内蒙古科技应用计划(2020GG0094)
内蒙古自治区高等学校科研项目(NJZY21321)
内蒙古自治区自然科学基金(2021LHMS06001)
内蒙古水利发展基金(NSK202109)
内蒙古自治区自然科学基金(2019MS03014)
内蒙古自治区科技重大专项(2019ZD016)
出版年
2024
NETL
NSTL
万方数据