Prediction of daily average amount of fine dust by tensorflow.
Pre-processing daily average amount of fine dust(pm10), bundling it into 7 days.
For the convolutional layer, the amount of fine dust for each region is mapped to the geographical map of Seoul, 9x8 matrix.
The data was extracted from Degree of Seoul Daily Average Air Pollution, provided by the Seoul Open Data Plaza.
Convolutional LSTM is similar as LSTM, but it calculate hidden states by convolutional operation.
Feed 5D Tensor (time_dim, batch_size, height, width, channel) to ConvLSTM Cell and it returns hidden unit and state.
There is prediction and forecasting models. Prediction model flatten an output of the last ConvLSTM Cell by 1x1 convolution and calculate MSE loss with the next day's fine dust data.
Forecasting model is the Stacked ConvLSTM Encoder-Decoder model. Stacked encoder summary the sequential data to the fixed-length vector and it is feeded to an initial state of the decoder model. Stacked decoder model generate stacked sequential data and it is flattened by 1x1 convolution. It produces 7 days prediction.
Model was converged.
In addition to the fine-dust, the accuracy can be improved by adding wind direction or date information.
Prediction
Forecaster
