Fig 1.
Basic structure of CNN model.
Fig 2.
BiLSTM model basic structure.
Fig 3.
Attention mechanism structure diagram.
Fig 4.
Raw data time series.
Table 1.
Variables included in the dataset.
Fig 5.
Pearson correlation analysis heat map of each variable.
Table 2.
Correlation and significance analysis results between PM2.5 and other variables.
Fig 6.
"Feature parameter-label" corresponding form.
Fig 7.
Multidimensional single-step prediction data sample generation process.
Fig 8.
Multi-dimensional and multi-step forecasting data sample generation process.
Fig 9.
CNN-BiLSTM-attention parallel structure model.
Fig 10.
CNN-BiLSTM-attention serial structure model.
Fig 11.
Changes in PM2.5 concentration.
Table 3.
Parallel structure model experimental results.
Table 4.
Serial structure model experimental results.
Table 5.
Model hyperparameter settings.
Fig 12.
PM2.5 concentration prediction results.
Table 6.
Evaluation index results of 10 groups of experiments.
Fig 13.
Comparison of the true and predicted values of each model.
Fig 14.
True and predicted values of different models.
Fig 15.
Comparison of the true and predicted values of each model.
Fig 16.
Long-term prediction results of PM2.5 concentration.
Table 7.
Long-term prediction and evaluation indicators of each model.
Fig 17.
Comparison of measured data and predicted value.