摘要
论文针对现有环境空气污染物预测方法大多是基于单一数据集和浅层神经网络,未能充分挖掘时间序列中潜藏的数据信息的问题,提出了一种基于门控循坏单元网络的空气污染物预测方法。首先,对时间序列缺失值进行设计填充算法;接着,设置监督实验,在批尺寸和训练步、训练优化算法、网络权值初始化和Dropout正则化四个方面进行参数调优;最后,进行了验证与分析,并与长短时记忆神经网络进行了参数对比。研究结果表明,与长短时记忆神经网络相比,门控循环单元网络不仅训练时间快,并且在预测性能上更为显著,是一种可行且有效的预测方法。
In this paper,aiming at the exitsing methods of ambient air pollutants prediction are based on a single data set and a shallow neural network,and the data information hidden in the time series can't be fully exploited.A time series prediction based on gated recurrent unit network is proposed.Firstly,a filling algorithm is designed for the missing values of time series.Then the supervised experiment is set up to adjust the parameters of batch size and training step,training optimization algorithm,network weight initialization and Dropout regularization,and the length and time are comparison of parameters of long short-term memory network.Finally,the verificaton and analysis are carried out,and the parameters are compared with the long-term memory time recurrent neural network.The research results show that compared with long-term memory time recurrent neural network,threshold loop unit network not only has a faster training time,but also has a more significant in air pollutant prediction performance,which is a feasible and effective prediction method.
作者
刘栩粼
谢崇波
LIU Xuin XIE Chongbo(Sichuan Vocational College of Information Technolog,Guangyuan 628000)
出处
《计算机与数字工程》
2024年第4期1257-1263,共7页
Computer & Digital Engineering
基金
四川信息职业技术学院青年科研基金项目(编号:2020C24)资助。
关键词
门控循环单元网络
时间递归神经网络
时间序列
深度学习
缺失值算法
gated recurrent unit network
time recurrent neural network
time series
deep learning
missing value algo-rithm
作者简介
刘栩粼,女,硕士研究生,助教,研究方向:无人机飞控、容错技术、软件技术;谢崇波,男,硕士研究生,助教,研究方向:数据挖掘及深度神经网络分析。
