Based on Air Quality Predictions with a Semi-Supervised Bidirectional LSTM Neural Network by Zhang et al.
Cite as: Zhang L., Liu P., Zhao L., Wang G., and Liu J. (2020). Air quality predictions with a semi-supervised bidirectional lstm neural network. Atmospheric Pollution Research, 12(1).
@article{2020Air,
title={Air quality predictions with a semi-supervised bidirectional LSTM neural network},
author={ Zhang, L. and Liu, P. and Zhao, L. and Wang, G. and Liu, J. },
journal={Atmospheric Pollution Research},
volume={12},
number={1},
year={2020},
}
our experiment envs (environment.yml) mainly contains：
CUDA 9.0
tensorflow 1.13.1 or later
keras 2.2.4 or later
emd-signal 1.0.0 (pip install EMD-signal)
all original PM2.5 data and EMD decomposed data are in folder ./data/ ;
all model results are in folder ./result/xxx
experiment step:
Step 1. Run EMD_decompose.py for original PM2.5 data decomposing.
Obtain EMD decompositions. (imf)
Step 2. Modify the config param in Config.py
VERSION: different experiment control.
IMF_NUM: experimental data selection in step 1
Step 3. Run Train.py for training the selected imf data.
Step 4. Repeat step 2 and 3 until all decompositions are trained.
Step 5. Modify the config param in Config.py
Run Predict.py for different imf.
Modify line 23 in Predicted.py (choose model param files)
Step 6. Aggregate all imf results to obtain final PM2.5 predictions.