inuyashacas / Air-quality-prediction_convLSTM

We propose to continuously predict the whole $station \times feature$ map in the next 1 to 48 hours with a convolutional LSTM (ConvLSTM, long short-term memory), which replaces the fully connected network in the classical LSTM with convolutional operations.

Air-quality-prediction

We propose to continuously predict the whole $station \times feature$ map in the next 1 to 48 hours with a non-stationary convolutional LSTM ( long short-term memory), which replaces the fully connected network in the classical LSTM with detrending and convolutional operations.

Prediction on consecutive four frames

Baseline

Prediction for seven cities of China, the Beijing, Shanghai, Chengdu, Wwuhan, Gguangzhou, Xi'an, and Nanjing.