摘要
随着社会迅速发展,空气污染对人类健康构成严重威胁。为有效预防,提出了基于频率域信息与双向长短期记忆(Frequency-Domain Information Bidirectional Long Short-Term Memory, FD-BiLSTM)神经网络的PM2.5浓度预测模型。利用离散余弦变换捕获频率特征,捕捉数据的周期性和趋势;通过BiLSTM模型预测结果,利用公开数据集对PM2.5浓度预测模型性能进行评估并验证。实验结果表明,多变量FD-BiLSTM模型能有效捕捉影响空气质量的复杂关系,可以实现更准确的PM2.5浓度预测。
With the rapid development of society,air pollution poses a serious threat to human health.For effective prevention,a PM2.5 concentration prediction model based on Frequency-Domain Information Bidirectional Long Short-Term Memory(FD-BiLSTM)network is proposed.Firstly,frequency features are captured using discrete cosine transform to capture periodicity and trend of the data.Then,the results are predicted by a BiLSTM model.Finally,the performance of PM2.5 concentration prediction model is evaluated and validated using publicly available datasets.Experimental results show that the multivariate FD-BiLSTM model can effectively capture complex relationships affecting air quality and can achieve more accurate PM2.5 concentration prediction.
作者
唐雪明
吴楠
TANG Xueming;WU Nan(School of Physics and Electronics,Nanning Normal University,Nanning 530199,China;School of Computer and Information Engineering,Nanning Normal University,Nanning 530199,China)
出处
《无线电通信技术》
2023年第6期1134-1141,共8页
Radio Communications Technology
基金
广西研究生教育创新计划项目(YCSW2023437)~~。
