基于改进时间卷积网络的空气质量预测研究  

Research on Air Quality Forecasting Based on Improved Temporal Convolutional Networks

作　　者：林涛 吉萌萌 付崇阁 程淑伟 LIN Tao;JI Meng-meng;FU Chong-ge;CHENG Shu-wei(School of Artificial Intelligence,Hebei University of Technology,Tianjin 300401,China)

机构地区：[1]河北工业大学人工智能与数据科学学院,天津300401

出　　处：《计算机仿真》2022年第10期451-456,501,共7页Computer Simulation

基　　金：国家自然科学基金资助项目(U20A20198);河北省重点研发计划项目新能源与智能电网技术创新专项(19214501D)。

摘　　要：针对空气质量数据包含的噪声较大、冗余因素过多而导致空气质量预测精度较低的问题,提出了一种收缩的时间卷积网络模型(Shrinking Temporal Convolutional Network, STCN)。模型利用时间卷积网络(Temporal Convolutional Network, TCN)的空洞因果卷积,保证较长的历史信息输入及未来信息的保密;利用深度残差收缩网络中特殊注意力机制和软阈值化的思想对TCN中的残差模块进行了改进,解决了因输入样本中的冗余信息不同导致的重要信息权重分散问题。实验结果表明,该方法能够有效地克服数据中噪声较大、冗余因素过多的问题,相较于LSTM、TCN等算法,该方法的准确率提高了1%~7%。This paper proposes a Shrinking Temporal Convolutional Network(STCN) model to address the problem of low air quality prediction accuracy caused by high noise and excessive redundancy factors in air quality data. The model uses the dilated causal convolution of the Temporal Convolutional Network(TCN) to ensure long historical information input and the confidentiality of future information;the special attention mechanism in the deep residual shrinkage network and the idea of soft thresholding are used to improve the residual module in TCN,which solves the problem of weight dispersion of important information caused by different redundant information in the input samples. The results of the experiment show that this method can effectively overcome the problems of large noise and excessive redundant factors in the data. Compared with algorithms such as LSTM and TCN,the accuracy of this method is improved by 1% to 7%.

关 键 词：空气质量预测 时间序列数据 时间卷积网络 软阈值化 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]