基于EMD和时空图神经网络的污染物浓度预测研究

Research on Air Pollution Concentration Prediction Model Based on Neural Network of EMD and Spatio-temporal Diagram

为了满足对大范围污染物浓度预测准确性和鲁棒性的要求,提出了一种基于EMD数据自增强的时空图神经网络预测模型。首先,使用encode-decode模型实现多步预测。其次使用EMD进行数据自增强将数据分解成若干线性数据,有助于提取数据间的隐藏特征。最后使用GNN+GRU时空图神经网络,同时从污染物的空间传输、时间周期的强自相关性及包含气象因素在内的邻域信息三个维度对污染物浓度进行预测。验证实验采用提出模型在真实数据集中预测PM_(2.5)和O_(3)在未来24、48和72小时的浓度,结果表明混合模型在预测精度中有明显提升。

In order to meet the requirements for the accuracy and robustness of the prediction of a wide range of pollutant concen⁃trations,a spatio-temporal graph neural network prediction model based on self-enhanced EMD data is proposed.First,the encodedecode model is adopted to achieve multi-step prediction.Secondly,EMD is introduced to decompose the data into several linear data,which helps to extract the hidden features between the data and increase the interpretability of the model.Finally,the GNN+GRU spatio-temporal graph neural network is used to predict the concentration of pollutants from the three dimensions of spatial transmis⁃sion of pollutants,strong autocorrelation of time periods,and critical domain knowledge including meteorological factors.The proposed hybrid model predicts the concentrations of PM_(2.5) and O_(3) in the next 24,48 and 72 hours in a large-scale real-world dataset.The re⁃sults show that the proposed model not only has a significant improvement in prediction accuracy,but also can take into account both peak prediction and trend changes.