%0 Journal Article
%T 基于ConvGRU深度学习网络模型的PM<sub>2.5</sub>浓度预测<br>PM<sub>2.5</sub> Concentration Prediction Based on ConvGRU Deep Learning Network Model
%A 文烨
%J Statistics and Applications
%P 71-78
%@ 2325-226X
%D 2024
%I Hans Publishing
%R 10.12677/SA.2024.131008
%X 针对于大气污染输送对PM<sub>2.5</sub>浓度预测影响的问题，本文以珠江三角洲作为研究区域，采用珠江三角洲PM<sub>2.5</sub>小时浓度数据作为研究数据，基于卷积操作和GRU模型构建了一种能够同时考虑时间依赖特征和空间特征的ConvGRU模型。并使用ConvGRU模型预测了珠海市PM<sub>2.5</sub>小时浓度，结果表明：ConvGRU模型预测PM<sub>2.5</sub>浓度和真实PM<sub>2.5</sub>浓度相关系数高达0.83，均方误差(MSE)、平均绝对误差(MAE)、平均绝对百分比误差(MAPR)、均方根误差(RMSE)均显著小于SVM和RF和ConvLSTM、LSTM和GRU。<br />
In response to the issue of predicting the impact of atmospheric pollution transport on PM<sub>2.5</sub> con-centrations, this paper takes the Pearl River Delta as the research area. Hourly PM<sub>2.5</sub> concentration data from the Pearl River Delta were used as the research dataset. A ConvGRU model, which com-bines convolutional operations and GRU (Gated Recurrent Unit), was developed to account for both temporal dependencies and spatial features. The ConvGRU model was employed to predict the hourly PM<sub>2.5</sub> concentrations in Zhuhai. The results showed that the ConvGRU model achieved a high correlation coefficient of 0.83 between the predicted PM<sub>2.5</sub> concentrations and the actual values. Moreover, it outperformed other models such as SVM, RF, ConvLSTM, LSTM, and GRU, with signifi-cantly lower values of mean squared error (MSE), mean absolute error (MAE), mean absolute per-centage error (MAPR), and root mean squared error (RMSE).
%K PM<sub>2.5</sub>，ConvGRU模型，GRU，ConvLSTM<br>PM<sub>2.5</sub>
%K ConvGRU Model
%K GRu
%K ConvLSTM
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=81029